# Analytical Studies Branch Research Paper SeriesIntergenerational Income Transmission: New Evidence from CanadaAnalytical Studies Branch Research Paper SeriesIntergenerational Income Transmission: New Evidence from Canada

## Archived Content

Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.

by Wen-Hao Chen and Yuri Ostrovsky, Social Analysis and Modelling Division, Statistics Canada
and Patrizio Piraino, University of Cape Town, South Africa

Release date: June 17, 2016

## Acknowledgements

The authors wish to thank Miles Corak for very detailed and useful comments on an earlier draft of this paper. The opinions expressed are those of the authors and do not necessarily reflect the views of Statistics Canada. Any errors are the sole responsibility of the authors.

## Abstract

This paper uses an updated version of a unique administrative dataset for Canada to test the impact of lifecycle earnings variation and errors-in-variables bias on estimates of the intergenerational income elasticity. The study finds higher levels of intergenerational earnings and income persistence compared with previous studies. Lifecycle bias in early estimates explains nearly two-thirds of the discrepancy, while errors-in-variables bias contributes to the remaining difference. The study shows that these biases have a smaller impact on daughters than on sons. The improved dataset is also used to re-examine nonlinear patterns in the intergenerational transmission of income in Canada. The findings suggest that limited mobility at the top of the distribution accounts for much of the average income persistence across generations, while mobility is found to be significantly higher among children born to low-income fathers. A comparison of the findings of this study with the international evidence reveals that nonlinear patterns in Canada are somewhat different from those observed in the United States and resemble the patterns found in Northern Europe.

JEL classification: J62, D31, D63

Keywords: earnings inequality, intergenerational mobility

## Executive summary

Comparative studies of intergenerational earnings and income mobility largely rank Canada as one of the most mobile countries among advanced economies, such as Denmark, Finland and Norway. The assertion that Canada is a highly mobile society is drawn from intergenerational income elasticity estimates reported in Corak and Heisz (1999). Corak and Heisz used data from the earlier version of the Intergenerational Income Database (IID), which tracked the income of Canadian youth only into their early thirties. Recent theoretical literature, however, suggests that the relationship between childrens' and parents' lifetime income may not be accurately estimated when children's income is not observed from their mid-careers—known as lifecycle bias.

The present study addresses this concern by re-examining the extent of intergenerational earnings and income mobility in Canada using the updated version of the IID, which tracks children well into their mid-forties, when mid-career income is observed. This information is essential for intergenerational analysis, as the literature shows that bias arising from lifecycle variation can be greatly mitigated by comparing fathers' and offspring's earnings near their mid-careers. Moreover, this paper also examines whether intergenerational mobility differs across the population. With nearly 250,000 observations, the study can differentiate the degree of intergenerational transmission across the full spectrum of the income distribution.

The empirical analysis in this study is based on Statistics Canada's IID, which was constructed from various tax records to link together children and their parents. The IID consists of youth aged 16 to 19 in 1982 whose tax records are linked to the tax records of their parents by means of the parents' and the children's Social Insurance Numbers and information from Statistics Canada's T1 Family File. The data provide more than 20 years of income history for both parents (1978 to 1999) and children (1986 to 2008) that allows for comparison of the income of children and parents when they were at the main stage of the lifecycle.

The results from the analysis suggest that Canada is still a mobile society, but not to the same extent as previously thought. The new estimate of the father–son earnings elasticity is about 0.32, which is noticeably higher than the values previously reported in the literature (which have been in the neighbourhood of 0.2): lifecycle bias alone explains about two-thirds of the discrepancy between the early estimates and the new result. The extent of intergenerational persistence tends to be greater when market income (i.e., the sum of earnings, self-employment income and asset income) is measured. This suggests that other mechanisms, such as transmission of jobs or entrepreneurial skills, may also be at work. Interestingly, the analysis also shows that the father–daughter elasticity is much less sensitive to these biases. Moreover, the paper documents a clear pattern of nonlinearity in the intergenerational transmission of earnings and income in Canada. In particular, the path to the top of the distribution appears to be quite challenging for sons born to low-income fathers. On the other hand, these same sons appear to have significant chances of moving into the middle class. Social institutions may help explain the latter findings. Finally, this paper demonstrates that the patterns of nonlinearity can be significantly misread when the lifecycle bias is not adequately addressed, especially over the upper part of the distribution.

## 1 Introduction

Comparative studies of intergenerational earnings and income mobility largely regard Canada as one of the most mobile countries among advanced economies (Björklund and Jäntti 2010; Corak 2013; Solon 2002). A popular chart—known as the "Great Gatsby Curve"—is often used to depict the relationship between income inequality and social mobility, and depicts Canada as one of the countries with the highest mobility, at levels similar to those observed for Denmark, Finland and Norway.Note 1 The assertion that Canada is a highly mobile society is drawn primarily from intergenerational income elasticity (IGE) estimates reported in Corak and Heisz (1999), as well as in Fortin and Lefebvre (1998). While these are careful studies that offer the best estimates of the IGE based on the data available at the time of their writing, recent literature suggests that such estimates may be subject to bias due, in particular, to lifecycle variation (Mazumder 2005; Haider and Solon 2006; Böhlmark and Lindquist 2006).

In the absence of data on lifetime earnings, IGE estimates may be affected by measurement error on both sides of the equation. Lifecycle bias may arise in a father–child analysis when a child's permanent earnings, proxied by yearly or short-run average earnings, are observed at a time of the child's working career when earnings do not closely correspond to their lifetime values. Haider and Solon (2006) showed that this type of bias usually results in lower estimates of the IGE when earnings from the early (or too late) stages of the working careers are used. Their finding is consistent with reviews of the literature showing that the smallest IGE estimates are most often found in studies where sons' earnings are observed at younger ages (Solon 1999). This is also evident in Corak et al. (2014) who find that the Canadian IGE estimate increases from 0.22 to 0.26 when sons' earnings are observed at age 35 rather than 30, suggesting that lifecycle effects should not be ignored in Canada as well.Note 2

In addition to lifecycle bias, the IGE estimates may also be affected by the classical errors-in-variables bias (right-hand-side measurement error), which occurs when fathers' lifetime earnings are not adequately measured. Corak and Heisz (1999), for instance, addressed this problem by using a five-year average instead of single-year earnings to approximate fathers' lifetime earnings. While such practice reduces the errors-in-variables bias, it may not eliminate it entirely if the number of years used in the calculation of the averages is insufficient. Mazumder (2005) showed that even estimates based on five-year average earnings are still subject to a significant errors-in-variable bias.

The primary objective of this study is thus to re-examine the extent of intergenerational earnings and income mobility in Canada in light of the estimation issues raised in the literature. Using longitudinal earnings data from a significantly higher number of years compared with previous research, the study highlights the sensitivity of IGE estimates to lifecycle and errors-in-variables bias. Importantly, the paper also revisits the issue of nonlinearity in the intergenerational transmission of earnings. Improved measures of fathers' and childrens' permanent earnings may have a different impact on the estimated intergenerational persistence at different parts of the distribution. Both polynomial and quantile regression specifications are employed to investigate this possibility. With nearly 250,000 observations, it is possible to examine differences in the degree of intergenerational mobility across the full spectrum of the income distribution.

The empirical analysis is based on an augmented version of the Intergenerational Income Database (IID), a high-quality dataset linking administrative records of parents and children in Canada. The data make it possible to introduce two essential novel elements compared with the earlier literature. First, the updated IID maintains the exceptionally large sample size of earlier versions, while offering a significantly longer panel of tax records for children, who are now observed from their late teen years well into their mid-forties. This information is essential for testing the impact of lifecycle bias on the IGE estimates, as the literature shows that this bias can be greatly mitigated by comparing fathers' with an offspring's earnings near their mid-careers (Grawe 2006; Haider and Solon 2006; Gouskova, Chiteji and Stafford 2010). Second, instead of being averaged over an arbitrarily defined calendar time period, income in this study is averaged over a specified age range. With up to 22 years of valid data, a father's lifetime income is approximated by averaging the income he received over the ages of 35 to 55. This approach should significantly reduce (if not eliminate) the attenuation bias arising from measurement error in fathers' permanent income.

The results from the analysis carried out in this study suggest that the extent of intergenerational earnings and income mobility in Canada was overestimated in the early studies. The new estimate of the father–son earnings elasticity is about 0.32, a result that is noticeably higher than the values previously reported in the literature (in the neighbourhood of 0.2). Failing to account for lifecycle bias explains about two-thirds of the difference between the current and previous estimates, while errors-in-variables bias contributes to another one-third of the discrepancy. Interestingly, the analysis performed in this study also shows that the father–daughter elasticity is much less sensitive to these biases. Finally, the study finds that the impact of such biases is more pronounced at the top of the income distribution, and documents a clear pattern of nonlinearity in the intergenerational transmission of earnings and income in Canada. In particular, the path to the top of the distribution appears to be quite challenging for sons born to low-income fathers. On the other hand, these same sons appear to have significant chances of moving into the middle class.

The remainder of the paper is organized as follows. Section 2 describes the data and the calculation of permanent earnings and income. Section 3 examines the sources of bias arising from imperfect measures of lifetime income and how these may affect estimates of the IGE. Section 4 presents the new estimates of intergenerational earnings and income mobility in Canada for both father–son and father–daughter pairs. Section 5 addresses the issue of nonlinearity, while Section 6 concludes.

## 2 Data and calculation of lifetime income

The empirical analysis in this study is based on matched parent–offspring tax records from an updated version of the IID file. A detailed description of the original IID file, which covered the period from 1978 to 1995, can be found in Corak and Heisz (1999). In essence, the IID file consists of three sub-samples of children aged 16 to 19—in (i) 1982, (ii) 1984 and (iii) 1986—whose tax records are linked to the tax records of their parents by means of the parents' and the childrens' Social Insurance Numbers and information from Statistics Canada's T1 Family File (T1FF). To improve coverage, child–parent pairs are drawn from the T1FF over all years from 1982 to 1986; where such links are available in multiple years, the earliest ones are retained. Once the child–parent link is established, it is possible to track children's and parents' annual earnings from 1978 to the latest available year (1995 in the original IID and 2008 in the updated file) by means of annual tax files (T1) and individuals' unique longitudinal identifiers based on their SINs.

One of the limitations of the original version of the IID was the relatively short number of adult years over which sons and daughters could be observed. For instance, even the oldest cohort of children—those born during the period from 1963 and 1966—could be observed only up to 29 to 32 years of age (in 1995). Observations on the other two cohorts of children were limited to even younger ages. The recently updated version of the IID has the same structure as the original file—same three cohorts of children—but it extends the sample period up to 2008. Hence, the children from the 1963-to-1966 cohort can now be observed up to 42 to 45 years of age. The child cohorts born during the periods from 1965 to 1968 and from 1967 to 1970 can now be observed up to 40 to 43 years of age and 38 to 41 of age, respectively. The analysis in this study focuses mainly on the 1963-to-1966 cohort of children and their linked fathers.

It should be mentioned that not all existing father–child pairs can be identified in the IID. First, a generational link cannot be established if a child, still living with his/her parents, did not file a tax return in any year from 1982 to 1986. Second, generational links cannot be established for children linked to families that had no records of fathers from 1982 to 1986. Finally, father–child links cannot be established for children whose records could not be linked to any family.Note 3 Along with Solon (1992) and Corak and Heisz (1999), this study uses only the oldest sons (daughters) when more than one son (daughter) is matched to the same father.

The analysis looks at three different income measures: earnings, market income and total income. Earnings are measured as a sum of wages from T4 slips issued by employers and other employment income, including tips, gratuities and directors' fees. Market income further includes rental income, self-employment income as well as asset income.Note 4 Total income refers to market income plus all government transfers, such as unemployment insurance benefits and pension benefits, but excluding taxes.Note 5 All monetary amounts are expressed in constant 2010 dollars.

### 2.1 An improved measure of lifetime earnings and income for fathers

It is well recognized in the literature that the use of current or single-year income as a proxy for permanent income can result in significant errors-in-variables biases in the estimation of the IGE (Solon 1999). A common remedy is to use multi-year averages to reduce the transitory component of income. However, in the absence of a full history of lifetime data, the number of years over which researchers can observe income is rather limited—usually three to five years. While the literature seems to agree that multi-year averages are better measures of permanent income than single-year records, there has been relatively little discussion on the exact number of years required for the computation of the averages. Is taking five-year averages sufficient to approximate for lifetime income? Using a unique (but rather small) set of linked survey and administrative data, Mazumder (2005) concluded that short-term proxies for fathers' permanent income may still be susceptible to bias as the variance of the transitory component of income varies considerably by age.

Moreover, the age at which fathers' income is averaged is sometimes overlooked in the literature. Corak and Heisz (1999), for instance, calculated fathers' permanent income by averaging their annual income over the period from 1978 to 1982. However, some fathers in the sample may have been too young or too old during the observed period and, therefore, the averages may not have properly captured their permanent income. Appendix Table 1 displays the age distribution of fathers in the IID. Indeed, nearly one-quarter of fathers in the sample were, arguably, either too young or too old from 1978 and 1982: about 5.3% of the fathers were aged 34 and under at the beginning of 1978, and nearly 18% of fathers were aged 56 or over by the end of 1982. As expected, the earnings of these fathers were significantly lower compared with those of their prime-age counterparts. Even within the prime-age group, there was some degree of variation: the mean earnings tended to be higher among fathers who were aged from 41 to 45 during the period from 1978 to 1982.

To improve the measure of fathers' lifetime income, their annual income at the age of 35 to 55 was averaged, conditional on their having had positive values ($500 and over in constant 2010 dollars) in at least 10 of these 21 years. The age restriction ensures that the income averages for fathers in the sample are calculated at a similar stage of their lifecycles and therefore are less affected by the larger transitory components typical of the early and late stages of individuals' working careers. Restricting the sample to fathers with 10 or more (non-successive) years of positive income further reduces variation driven by few high- or low-income years within the prime age. Note that, for each father in the IID, earnings and income data were available from 1978 to 1999. This implies that different cohorts of fathers will have different years over which their earnings can be averaged (see Appendix Table 2 for an illustration).Note 6 In order to have at least 10 positive annual records that satisfy both the age (35 to 55) and the calendar time (1978 to 1999) restrictions, only fathers born in the years from 1932 to 1955 could be included.Note 7 It is important to point out that the proposed measure is regarded as an improved proxy for lifetime income, rather than its "true" measure. The final sample consisted of 356,321 fathers (with positive lifetime earnings) that can be matched to childrens' records.Note 8 Of these fathers, 56.4% were from the 1932-to-1938 cohorts, and 40.2% were from the 1939-to-1945 cohorts. Only 3.4% were born during the period from 1946 to 1955. ## 3 Empirical testing for lifecycle bias and errors-in-variables bias Using U.S. data, Haider and Solon (2006) showed that the bias arising from lifecycle variation could be greatly reduced if children' earnings were measured at their mid-careers. However, determining the age that minimizes the lifecycle bias is an empirical task. It is important to note for the purposes of this study that such age may be country-specific. In this section, Haider and Solon's generalized errors-in-variables model is applied to IID data. It should be emphasized that the purpose of this exercise was not to formally examine the association between annual and lifetime earnings, as was done by Haider and Solon (2006), since this would require full lifelong-earnings histories, which were not available in the IID.Note 9 In addition, the analysis in this study was conditional on the sample of men who were intergenerationally linked at a certain point in their lives, which may not be comparable to the broader samples used in studies that have focused mainly on lifecycle earnings variation. As a consequence, this section is intended solely to illustrate the bias in IGE estimates, when the lifetime income of fathers and offspring cannot be adequately measured. Following the literature standards, a simple model describing the relationship between fathers' and children's incomes can be written as: $Y i s =α+β Y i f + ε i (1) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGzbWdamaaDaaaleaapeGaamyAaaWdaeaapeGaam4Caaaakiab g2da9iabeg7aHjabgUcaRiabek7aIjaadMfapaWaa0baaSqaa8qaca WGPbaapaqaa8qacaWGMbaaaOGaey4kaSIaeqyTdu2damaaBaaaleaa peGaamyAaaWdaeqaaOGaaGzbVlaaywW7caaMf8Uaaiikaiaaigdaca GGPaaaaa@4C3D@$ where ${Y}_{i}^{s}\left({Y}_{i}^{f}\right)$ is the log of child's (father's) lifetime income, ${\epsilon }_{i}$ is a random error uncorrelated with ${Y}_{i}^{f}$, and $\beta$ captures the IGE. Since permanent income is usually not available for either fathers or children, the attempt to measure this variable by yearly data or even multi-year averages—which is common in the literature—could result in biased estimates of the IGE as a result of measurement errors on both sides of the equation. To illustrate this, the analysis follows Haider and Solon (2006) and defines the left-hand-side income measure (children's earnings in this study) as: $Y it = λ t Y i + u it , (2) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGzbWdamaaBaaaleaapeGaamyAaiaadshaa8aabeaak8qacqGH 9aqpcqaH7oaBpaWaaSbaaSqaa8qacaWG0baapaqabaGcpeGaamywa8 aadaWgaaWcbaWdbiaadMgaa8aabeaak8qacqGHRaWkcaWG1bWdamaa BaaaleaapeGaamyAaiaadshaa8aabeaakiaayIW7caaMi8Uaaiilai aaywW7caaMf8UaaGzbVlaacIcacaaIYaGaaiykaaaa@4E6F@$ where ${Y}_{i}$ is the lifetime income proxied by ${Y}_{it}$ (annual earnings at age $t$), ${u}_{it}$ is a random disturbance uncorrelated with ${Y}_{i}$ and ${\epsilon }_{i}$, and ${\lambda }_{t}$ is the slope coefficient in the so-called "forward regression" of ${Y}_{it}$ on ${Y}_{i}$. Suppose one wishes to estimate (1). Substituting (2) into (1) gives $Y it =a+ λ t β Y i f +( λ ε i + u it ) . (3) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGzbWdamaaBaaaleaapeGaamyAaiaadshaa8aabeaak8qacqGH 9aqpcaWGHbGaey4kaSIaeq4UdW2damaaBaaaleaapeGaamiDaaWdae qaaOWdbiabek7aIjaadMfapaWaa0baaSqaa8qacaWGPbaapaqaa8qa caWGMbaaaOGaey4kaSYaaeWaa8aabaWdbiabeU7aSjabew7aL9aada WgaaWcbaWdbiaadMgaa8aabeaak8qacqGHRaWkcaWG1bWdamaaBaaa leaapeGaamyAaiaadshaa8aabeaaaOWdbiaawIcacaGLPaaacaaMi8 UaaGjcVlaac6cacaaMf8UaaGzbVlaaywW7caGGOaGaaG4maiaacMca aaa@5A1E@$ Therefore, ${\lambda }_{t}\beta$ (instead of $\beta$) is the probability limit of the estimated coefficient on ${Y}_{i}^{f}$. As a result, the ordinary least squares (OLS) estimator of (1) is consistent only when ${\lambda }_{t}=1$. A lifecycle bias thus arises when ${\lambda }_{t}\ne 1$, and this result may vary with the age at which incomes are observed. Haider and Solon (2006) demonstrated this for the United States. ${\lambda }_{t}$ profiles vary notably across the lifecycle: the profile begins at 0.24 at age 19, rises to about 1 at age 32, and declines towards the end of the working career. Chart 1 presents estimates of ${\lambda }_{t}$ using the IID data described above. They are slope coefficients in the forward regression of log annual earnings, at age $t$, on log lifetime earnings (Equation 2). As explained above, lifetime earnings are calculated by taking average earnings over the ages of 35 to 55 (conditional on positive values for at least 10 years). This proxy may be considered an upper-bound estimate of lifetime earnings since it excludes low-earnings years (i.e., early and late parts of the lifecycle). Nonetheless, to the extent that the prime-age earnings are more representative of the lifetime values (Haider and Solon 2006; Böhlmark and Lindquist 2006), the coefficients from the forward regressions can provide some guidance on the possible presence of lifecycle bias in previous estimates of the IGE in Canada. The results in Chart 1 indicate that ${\lambda }_{t}$ does not equal 1 throughout the lifecycle. This value begins at 0.47 at age 25, increases gradually and reaches unity in the early forties. It continues to increase to about 1.14 at age 52, and falls back to below 1 after age 56. An important implication of Chart 1 is that the estimates of the IGE could be significantly underestimated when sons' earnings are observed at younger ages. This finding appears to justify the intention to revisit the previous Canadian estimates. Chart 1 also suggests that any bias arising from lifecycle variation may be mitigated when sons' earnings are measured somewhere around the late thirties or early forties (taking into account the fact that the permanent income measure in this study may overestimate lifetime earnings). Data table for Chart 1 Data table for Chart 1 Estimate of ${\lambda }_{t}$ (lifecycle bias) Table summary This table displays the data for Chart 1. The information is grouped by Age (appearing as row headers), Forward regression of log annual earnings on log lifetime earnings, 95% confidence interval (lower) and 95% confidence interval (upper), calculated using coefficient units of measure (appearing as column headers). Age Forward regression of log annual earnings on log lifetime earnings 95% confidence interval (lower) 95% confidence interval (upper) coefficient 25 0.4733 0.3466 0.5999 26 0.4837 0.3932 0.5742 27 0.5491 0.4850 0.6131 28 0.6277 0.5807 0.6748 29 0.6231 0.5868 0.6595 30 0.6617 0.6346 0.6888 31 0.6755 0.6541 0.6969 32 0.7073 0.6911 0.7234 33 0.7390 0.7260 0.7520 34 0.7565 0.7462 0.7668 35 0.8215 0.8135 0.8294 36 0.8508 0.8443 0.8572 37 0.8697 0.8644 0.8749 38 0.8989 0.8943 0.9034 39 0.9161 0.9121 0.9201 40 0.9407 0.9372 0.9443 41 0.9617 0.9585 0.9650 42 0.9824 0.9794 0.9853 43 1.0028 1.0001 1.0055 44 1.0213 1.0188 1.0239 45 1.0354 1.0330 1.0379 46 1.0523 1.0499 1.0546 47 1.0730 1.0706 1.0753 48 1.0942 1.0918 1.0966 49 1.1134 1.1109 1.1159 50 1.1309 1.1283 1.1336 51 1.1385 1.1357 1.1413 52 1.1402 1.1372 1.1432 53 1.1338 1.1306 1.1371 54 1.1239 1.1202 1.1276 55 1.0920 1.0878 1.0963 56 0.9928 0.9875 0.9981 57 0.9334 0.9273 0.9395 58 0.8757 0.8688 0.8827 59 0.8202 0.8122 0.8283 60 0.7717 0.7622 0.7812 Haider and Solon (2006) also addressed the right-hand-side measurement error. The unobserved regressor, ${Y}_{i}$ (fathers' lifetime earnings, in this case), is proxied by annual earnings at age $t$: $Y i = θ t Y it + v it . (4) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGzbWdamaaBaaaleaapeGaamyAaaWdaeqaaOWdbiabg2da9iab eI7aX9aadaWgaaWcbaWdbiaadshaa8aabeaak8qacaWGzbWdamaaBa aaleaapeGaamyAaiaadshaa8aabeaak8qacqGHRaWkcaWG2bWdamaa BaaaleaapeGaamyAaiaadshaa8aabeaakiaac6cacaaMf8UaaGzbVl aaywW7caaMf8UaaGzbVlaacIcacaaI0aGaaiykaaaa@4E70@$ Equation (4) is called the "reverse regression" of ${Y}_{i}$ on , and the probability limit of the estimated slope coefficient is $plim β ^ = cov( Y it f , Y i s ) var( Y it f ) = θ t β , (5) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaaeaaaaaa aaa8qacaWFWbGaa8hBaiaa=LgacaWFTbGaaGPaVlqbek7aIzaajaGa aGPaVlabg2da9maalaaapaqaa8qacaWFJbGaa83Baiaa=zhadaqada WdaeaapeGaamywa8aadaqhaaWcbaWdbiaadMgacaWG0baapaqaa8qa caWGMbaaaOGaaiilaiaadMfapaWaa0baaSqaa8qacaWGPbaapaqaa8 qacaWGZbaaaaGccaGLOaGaayzkaaaapaqaaKqzagWdbiaa=zhacaWF HbGaa8NCaOWaaeWaa8aabaWdbiaadMfapaWaa0baaSqaa8qacaWGPb GaamiDaaWdaeaapeGaamOzaaaaaOGaayjkaiaawMcaaaaacqGH9aqp cqaH4oqCpaWaaSbaaSqaa8qacaWG0baapaqabaGcpeGaeqOSdiMaaG jcVlaayIW7caGGSaGaaGzbVlaaywW7caaMf8UaaGzbVlaaywW7caGG OaGaaGynaiaacMcaaaa@697E@$ where $θ t = cov( Y it f , Y i f ) var( Y it f ) = λ t var( Y i f ) λ t 2 var( Y i f )+var( v it ) . (6) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacqaH4oqCpaWaaSbaaSqaa8qacaWG0baapaqabaGcpeGaeyypa0Za aSaaa8aabaacbaWdbiaa=ngacaWFVbGaa8NDamaabmaapaqaa8qaca WGzbWdamaaDaaaleaapeGaamyAaiaadshaa8aabaWdbiaadAgaaaGc caGGSaGaamywa8aadaqhaaWcbaWdbiaadMgaa8aabaWdbiaadAgaaa aakiaawIcacaGLPaaaa8aabaWdbiaa=zhacaWFHbGaa8NCamaabmaa paqaa8qacaWGzbWdamaaDaaaleaapeGaamyAaiaadshaa8aabaWdbi aadAgaaaaakiaawIcacaGLPaaaaaGaeyypa0ZaaSaaa8aabaWdbiab eU7aS9aadaWgaaWcbaWdbiaadshaa8aabeaakiaa=zhapeGaa8xyai aa=jhadaqadaWdaeaapeGaamywa8aadaqhaaWcbaWdbiaadMgaa8aa baWdbiaadAgaaaaakiaawIcacaGLPaaaa8aabaWdbiabeU7aS9aada qhaaWcbaWdbiaadshaa8aabaWdbiaaikdaaaGcpaGaa8NDa8qacaWF HbGaa8NCamaabmaapaqaa8qacaWGzbWdamaaDaaaleaapeGaamyAaa WdaeaapeGaamOzaaaaaOGaayjkaiaawMcaaiabgUcaRiaa=zhacaWF HbGaa8NCamaabmaapaqaa8qacaWG2bWdamaaBaaaleaapeGaamyAai aadshaa8aabeaaaOWdbiaawIcacaGLPaaaaaGaaGjcVlaayIW7caaM i8UaaGjcVlaac6cacaaMf8UaaGzbVlaaywW7caaMf8UaaGzbVlaacI cacaaI1aGaaiykaaaa@80D6@$ When ${\lambda }_{t}=1$${\theta }_{t}$ is simply the textbook case of attenuation bias. However, ${\theta }_{t}$ will also depend on the value of ${\lambda }_{t}$. Haider and Solon (2006) argued that, in rare cases, ${\theta }_{t}$ can turn out to be an amplification, rather than attenuation, bias. The measure of lifetime earnings is used to estimate the trajectories of ${\theta }_{t}$ for Canadian men (Chart 2). As expected, using annual earnings to approximate lifetime values on the right-hand side of the regression results in a significant attenuation bias (solid line). The bias is especially pronounced $\left({\theta }_{t}=0.25\right)$ when current earnings are observed early or late in the lifecycle, and remains large (about 0.60) when earnings are measured at the mid-career years. These findings are in line with Björklund (1993) and Böhlmark and Lindquist (2006) for Sweden, as well as with Haider and Solon (2006) for the United States. In practice, researchers often use multi-year averages instead of yearly earnings as a proxy for lifetime earnings. To what extent do such practices reduce the errors-in-variables bias? To answer this question, the reverse regressions are re-estimated with current annual earnings replaced by mean annual earnings within a five-year window, centered at any given age.Note 10 Chart 2 (dashed line) reveals that the attenuation bias can be significantly mitigated when, instead of annual earnings, five-year averages are used to proxy lifetime earnings. The estimated ${\theta }_{t}$ can be as high as 0.85 when mid-career multi-year earnings averages are used. Chart 2 also supports Mazumder (2005), who argued that even estimates based on five-year averages are still subject to non-negligible errors-in-variables bias. All in all, results from both Charts 1 and 2 seem to suggest that the Canadian IGE may have been underestimated in previous studies. Data table for Chart 2 Data table for Chart 2 Estimate of ${\theta }_{t}$ (attenuation bias) Table summary This table displays the data for Chart 2. The information is grouped by Age (appearing as row headers), Reverse regression of lifetime earnings on annual earnings, 95% confidence intervals (lower), 95% confidence intervals (upper) and Reverse regression of lifetime earnings on "mean" annual earnings (that is to say, average over Plus-Minus 2 years at each age), calculated using coefficient units of measure (appearing as column headers). Age Reverse regression of lifetime earnings on annual earnings 95% confidence intervals (lower) 95% confidence intervals (upper) Reverse regression of lifetime earnings on "mean" annual earnings (i.e., average over ± 2 years at each age) coefficient 25 0.2894 0.2120 0.3669 0.4913 26 0.3095 0.2516 0.3674 0.5948 27 0.3822 0.3377 0.4268 0.6611 28 0.4198 0.3884 0.4513 0.6695 29 0.4218 0.3972 0.4464 0.7253 30 0.4630 0.4440 0.4820 0.7411 31 0.4679 0.4531 0.4828 0.7631 32 0.5038 0.4923 0.5154 0.7735 33 0.5089 0.5000 0.5178 0.7879 34 0.5136 0.5065 0.5206 0.8104 35 0.5463 0.5410 0.5516 0.8268 36 0.5589 0.5546 0.5631 0.8337 37 0.5836 0.5801 0.5871 0.8493 38 0.5867 0.5837 0.5897 0.8523 39 0.5973 0.5947 0.5999 0.8545 40 0.6006 0.5984 0.6029 0.8571 41 0.6050 0.6029 0.6070 0.8586 42 0.6122 0.6103 0.6140 0.8562 43 0.6158 0.6141 0.6175 0.8570 44 0.6199 0.6184 0.6215 0.8563 45 0.6209 0.6195 0.6224 0.8537 46 0.6214 0.6200 0.6228 0.8502 47 0.6157 0.6143 0.6170 0.8411 48 0.6069 0.6056 0.6083 0.8281 49 0.5886 0.5873 0.5899 0.8111 50 0.5660 0.5647 0.5673 0.7900 51 0.5446 0.5433 0.5459 0.7670 52 0.5174 0.5160 0.5187 0.7398 53 0.4869 0.4855 0.4883 0.7069 54 0.4432 0.4418 0.4447 0.6680 55 0.3888 0.3873 0.3903 0.6227 56 0.3042 0.3026 0.3058 0.5704 57 0.2604 0.2587 0.2621 0.5073 58 0.2274 0.2256 0.2292 0.4339 59 0.1967 0.1948 0.1987 Note ...: not applicable 60 0.1662 0.1642 0.1683 Note ...: not applicable ### 3.1 International comparison of bias profiles In this subsection, the estimated Canadian bias profiles are compared with those of other countries. The comparison may shed some light on the sources of cross-country differences in the intergenerational transmission of earnings. A few studies have investigated the association between current and lifetime income. Haider and Solon (2006) were the first to offer empirical estimates of ${\lambda }_{t}$ and ${\theta }_{t}$ based on nearly career-long earnings histories from a U.S. panel. Two other papers known to the authors have been able to replicate the approach followed by Haider and Solon for other countries: these are Böhlmark and Lindquist (2006) for Sweden, and Brenner (2010) for Germany, using data from Swedish tax records and the German Socio-Economic Panel, respectively.Note 11 The estimates of ${\lambda }_{t}$ and ${\theta }_{t}$ from these studies are shown in Panel A and Panel B of Chart 3, respectively. To improve comparability, Canadian men are restricted to those born during the period from 1939 to 1945—a cohort with an age range similar to that of the German and Swedish cohorts, but still about a decade younger than the American cohort. The Canadian profile is shorter because only up to 21 years of earnings data are available in the IID. In general, with respect to the estimated ${\lambda }_{t}$, all four studies show that the textbook scenario of ${\lambda }_{t}=1$ throughout the lifecycle does not apply. As a consequence, using annual earnings from early career stage to proxy for lifetime earnings, as the dependent variable, would lead to an attenuation bias. There is also a similarity in terms of where (or at what age) the lifecycle bias is minimized. The estimate of ${\lambda }_{t}$ seems to approach 1 at the age of 40 for Canadian men, but does so at a somewhat younger age—at around mid-thirties—for those of the other three countries. This may be due to the fact that the measure of lifetime earnings for Canada in this study is overestimated by definition (i.e., because individuals' earnings are averaged over their prime age). However, 95% confidence intervals around these profiles do overlap over the mid-career, at age 40 to 41. After 41, the bias profiles for each country become more distinct. For the United States, the lifetime earnings are underestimated from the mid-career on, while the opposite seems to be the case for Germany, where ${\lambda }_{t}$ continues to grow until the age of 44 and remains above unity throughout the rest of the career. The Canadian and Swedish profiles are somewhat similar in the sense that the estimated ${\lambda }_{t}$ in both countries peaks at around age 50 and declines thereafter, showing a more pronounced drop for Canadian men (which, again, may be due to the particular measure used here). These patterns are insightful for international comparisons of the intergenerational transmission of earnings. First, even when sample and methodological differences are accounted for, cross-national variation in IGE estimates may still arise, as a result of differences in the age at which sons' earnings are observed, everything else being equal.Note 12 For instance, using sons' earnings corresponding to the ages from the mid-forties to the mid-fifties tends to induce an attenuation bias in the United States, but an amplification bias in the other three countries. On the other hand, lifecycle bias seems to be greatly reduced when sons' earnings are measured around the age of 40 in all countries. For the trajectory of ${\theta }_{t}$ (Panel B), all countries exhibit a similar inverted-U-shaped profile over the lifecycle. Using current earnings to proxy for lifetime earnings as the independent variable will lead to downward bias when earnings from early and late stages of the working careers are used. In all cases, the bias is smaller around the mid-career, but ${\theta }_{t}$ remains far below unity. This suggests that the bias arising from the right-hand-side measurement error cannot be eliminated completely at any stage of the working life. It is interesting to note that, since ${\lambda }_{t}$ is approximately 1 for all four countries at about age 40, ${\theta }_{40}$ captures the attenuation bias due to classical measurement error. Again, cross-national differences in ${\theta }_{t}$ are more visible at the beginning and end of the working career. Data table for group Chart 3 Data table for group Chart 3 Cross-national comparison of bias profiles for men Table summary This table displays the results of Cross-national comparison of bias profiles for men. The information is grouped by Age (earnings are measured) (appearing as row headers), Estimate of ${\lambda }_{t}$, Estimate of ${\theta }_{t}$, Canadian men 1939 to 1945, American men 1931 to 1933, German men 1939 to 1944 and Swedish men 1939 to 1943, calculated using coefficient units of measure (appearing as column headers). Age (earnings are measured) Estimate of ${\lambda }_{t}$ Estimate of ${\theta }_{t}$ Canadian men 1939 to 1945 American men 1931 to 1933 German men 1939 to 1944 Swedish men 1939 to 1943 Canadian men 1939 to 1945 American men 1931 to 1933 German men 1939 to 1944 Swedish men 1939 to 1943 coefficient 20 Note ...: not applicable 0.251 -0.055 0.020 Note ...: not applicable 0.199 -0.016 0.020 21 Note ...: not applicable 0.171 0.013 -0.050 Note ...: not applicable 0.094 0.003 -0.010 22 Note ...: not applicable 0.284 0.139 0.160 Note ...: not applicable 0.144 0.038 0.050 23 Note ...: not applicable 0.317 0.215 0.165 Note ...: not applicable 0.258 0.087 0.050 24 Note ...: not applicable 0.405 0.278 0.280 Note ...: not applicable 0.352 0.106 0.090 25 Note ...: not applicable 0.432 0.400 0.370 Note ...: not applicable 0.652 0.211 0.110 26 Note ...: not applicable 0.541 0.457 0.560 Note ...: not applicable 0.620 0.282 0.325 27 Note ...: not applicable 0.508 0.537 0.670 Note ...: not applicable 0.696 0.359 0.455 28 Note ...: not applicable 0.643 0.643 0.720 Note ...: not applicable 0.684 0.371 0.410 29 Note ...: not applicable 0.760 0.683 0.780 Note ...: not applicable 0.629 0.412 0.437 30 Note ...: not applicable 0.835 0.738 0.810 Note ...: not applicable 0.630 0.460 0.370 31 Note ...: not applicable 0.945 0.817 0.880 Note ...: not applicable 0.644 0.472 0.400 32 Note ...: not applicable 0.993 0.883 0.840 Note ...: not applicable 0.625 0.496 0.490 33 0.6745 1.270 0.905 0.950 0.5209 0.527 0.487 0.490 34 0.6924 0.869 0.981 1.040 0.5389 0.756 0.512 0.440 35 0.7748 0.947 0.984 1.000 0.5558 0.700 0.576 0.500 36 0.8121 0.812 0.985 0.820 0.5694 0.803 0.567 0.550 37 0.8391 0.981 1.028 0.865 0.5909 0.688 0.580 0.600 38 0.8799 1.062 0.950 0.890 0.5834 0.640 0.632 0.508 39 0.9024 1.170 1.073 0.900 0.5943 0.573 0.588 0.504 40 0.9494 1.144 1.023 1.000 0.5891 0.578 0.649 0.500 41 0.9945 0.948 1.104 1.180 0.5837 0.630 0.585 0.385 42 1.0359 0.821 1.202 1.200 0.5833 0.700 0.561 0.425 43 1.0696 0.818 1.226 1.070 0.5800 0.663 0.570 0.500 44 1.0849 0.860 1.316 1.030 0.5812 0.667 0.531 0.580 45 1.0931 0.860 1.282 1.032 0.5827 0.642 0.557 0.482 46 1.1078 0.904 1.267 1.011 0.5834 0.568 0.578 0.565 47 1.1237 0.764 1.266 1.010 0.5680 0.628 0.569 0.410 48 1.1342 0.722 1.291 1.100 0.5489 0.594 0.578 0.480 49 1.1495 0.725 1.248 1.200 0.5281 0.584 0.602 0.385 50 1.1626 0.756 1.263 1.320 0.5043 0.519 0.552 0.350 51 1.1734 0.762 1.258 1.220 0.4792 0.512 0.569 0.360 52 1.1676 0.788 1.258 1.150 0.4416 0.484 0.554 0.412 53 1.1502 0.776 1.170 1.150 0.4133 0.477 0.561 0.460 54 1.1248 0.770 1.199 1.000 0.3776 0.456 0.519 0.365 55 1.0740 0.776 1.114 1.140 0.3309 0.429 0.481 0.495 56 0.9651 0.799 1.234 1.190 0.2597 0.423 0.471 0.475 57 0.8995 0.763 1.237 1.130 0.2229 0.379 0.440 0.460 58 0.8371 0.723 1.232 1.180 0.1963 0.368 0.380 0.460 59 0.7891 0.704 1.225 1.020 0.1754 0.315 0.354 0.510 60 0.7371 Note ...: not applicable Note ...: not applicable Note ...: not applicable 0.1521 Note ...: not applicable Note ...: not applicable Note ...: not applicable ## 4 New evidence of intergenerational earnings and income mobility in Canada To what extent are the estimates of the intergenerational elasticity in Canada affected by the use of inadequate proxies for both fathers' and sons' lifetime earnings? To answer this question, two different scenarios using alternative measures of fathers' lifetime earnings are presented in this part of the analysis. To compare the results from this study to previous findings, the first scenario follows Corak and Heisz (1999) and defines fathers' earnings as five-year averages over the period from 1978 to 1982. The solid line in Chart 4 presents estimates of the IGE obtained by running an OLS regression defined by scenario 1 for different ages for which sons' earnings are observed. Overall, the estimated IGE follows a concave trajectory, reflecting lifecycle differences in the magnitude of the bias from left-hand-side measurement error. When sons' earnings are observed at around age 30, as in Corak and Heisz (1999), the model produces an estimate of $\beta$ (IGE) that is nearly identical to the one in their study—about 0.227.Note 13 The estimated elasticity continues to increase as sons' earnings are measured at older ages. It rises to 0.29 when earnings are observed in the early forties (where ${\lambda }_{t}$ approaches 1), indicating a higher degree of intergenerational earnings persistence. Data table for Chart 4 Data table for Chart 4 Estimates of $\beta$ (IGE) by the age of sons when earnings are observed — Two scenarios for measuring fathers' lifetime earnings Table summary This table displays the data for Chart 4. The information is grouped by Age of sons (appearing as row headers), Scenario (1): Mean over 1978 to 1982 (Corak and Heisz 1999) and Scenario (2): Lifetime (mean over ages 35 to 55), calculated using coefficient units of measure (appearing as column headers). Age of sons Scenario (1): Mean over 1978 to 1982 (Corak and Heisz 1999) Scenario (2): Lifetime (mean over ages 35 to 55) coefficient 20 0.016 0.042 21 0.008 0.042 22 0.036 0.071 23 0.040 0.080 24 0.083 0.126 25 0.129 0.172 26 0.164 0.208 27 0.188 0.232 28 0.204 0.251 29 0.218 0.258 30 0.226 0.272 31 0.236 0.284 32 0.248 0.292 33 0.258 0.298 34 0.268 0.309 35 0.273 0.311 36 0.277 0.319 37 0.281 0.319 38 0.281 0.312 39 0.284 0.318 40 0.286 0.321 41 0.290 0.322 42 0.291 0.322 43 0.292 0.320 44 0.292 0.316 45 0.286 0.323 In addition to lifecycle bias, estimates of the IGE also depend on how fathers' lifetime earnings are calculated. Scenario 2 uses the measure for fathers' lifetime earnings introduced in this study in an attempt to minimize the right-hand-side measurement error. In general, scenario 2 (dashed line) produces higher earnings elasticity compared with the first scenario. The estimated $\beta$ now increases by another 3 points, to 0.32, when sons' earnings are observed in their mid-careers. Chart 4 confirms that both left-hand- and right-hand-side measurement errors can produce a downward bias in the IGE estimates. In particular, the left-hand side lifecycle bias can be substantial when the estimated IGEs are based on sons' earnings from the very early years of their working careers. ### 4.1 Are Canadians as mobile as previously suggested? Table 1 presents new estimates of the IGE for Canada. Again, different scenarios for measuring lifetime earnings of fathers and sons are offered in order to grasp the possible impact of left-hand- and right-hand-side measurement errors on the estimated coefficients. The baseline scenario (scenario 1) uses a definition of lifetime earnings similar to that used in Corak and Heisz (1999); scenario 2 mitigates left-hand-side lifecycle bias by observing sons' earnings at age 40; scenario 3 further reduces the right-hand-side attenuation bias by using fathers' earnings averaged over their prime age (from 35 to 55 years); and the last scenario—the preferred scenario in the context of this study—improves the precision of sons' lifetime earnings with the use of five-year averages around their mid-careers (i.e., ages 38 to 42). In addition to earnings, the results are presented for both market and total income (before taxes, after transfers). With respect to scenarios 3 and 4, the father–son intergenerational earnings elasticity in Canada is about 0.32—about 46% (or 10 points) higher than the commonly quoted estimate of 0.22 from Corak and Heisz (1999). Both sources of bias (lifecycle and errors-in-variables) contribute to the underestimation in the early studies. In general, failing to account for the lifecycle bias explains about two-thirds of the difference between the current and previous estimates, while the use of a less accurate proxy for fathers' lifetime earnings contributes to another one-third of the discrepancy. Comparing scenario 4 with scenario 3 in Table 1 also suggests that, when sons' earnings are observed at the "right" age (i.e., around their forties), replacing them by five-year averages does not significantly change the estimate of the IGE. This may suggest that there is a lesser need to approximate sons' lifetime earnings with multi-year averages as long as earnings data at their mid-careers can be observed. Table 1 Estimates of intergenerational earnings and income elasticity, fathers and sons Table summary This table displays the results of Estimates of intergenerational earnings and income elasticity. The information is grouped by Proxies for lifetime incomes (appearing as row headers), $\beta$ (fathers and sons), Earnings, Market income and Total income, calculated using coefficient and standard error units of measure (appearing as column headers). Proxies for lifetime incomes $\beta$ (fathers and sons) EarningsTable 1 - Note 1 Market incomeTable 1 - Note 2 Total incomeTable 1 - Note 3 coefficient standard error coefficient standard error coefficient standard error Scenario 1 Sons: at age 30 Fathers: mean over 1978 to 1982 0.227 0.003 0.230 0.003 0.222 0.002 Scenario 2 Sons: at age 40 Fathers: mean over 1978 to 1982 0.287 0.004 0.301 0.003 0.317 0.003 Scenario 3 Sons: at age 40 Fathers: mean over ages 35 to 55Table 1 - Note 4 0.321 0.004 0.349 0.003 0.359 0.003 Scenario 4, preferred Sons: mean over ages 38 to 42Table 1 - Note 5 Fathers: mean over ages 35 to 55Note 4 0.318 0.004 0.343 0.003 0.359 0.003 ### 4.2 Market and total income mobility In addition to earnings, the IID also includes other income sources that allow researchers to examine the intergenerational transmission of market as well as total income. Understanding income mobility can offer additional insight into transmission mechanisms across generations. Previous research, for instance, has shown clear evidence of the intergenerational transmission of jobs (Kramarz and Skans 2014), self-employment (Lentz and Laband 1990; Dunn and Holtz-Eakin 2000; Sorensen 2004), chief executive officer positions (Pérez-González 2006), liberal professions (Aina and Nicoletti 2013), and employers (Corak and Piraino 2011). Aside from financial transfers, these articles also emphasize the importance of other transmission mechanisms, such as transmission of human capital, networking and, in some cases, nepotism. Dunn and Holtz-Eakin (2000), for example, pointed out that parents' own entrepreneurial experience and business success has a significant effect on the propensity of becoming self-employed. This literature suggests that intergenerational persistence should be higher for market income, since this includes self-employment and asset income. The result in Table 1 confirms this expectation. The sample here further includes father–son pairs who have positive self-employment or asset income. The intergenerational elasticity for fathers and sons now increases by another 3 points to about 0.35 in the preferred model. It is also noteworthy that estimates based on scenario 1 significantly underestimate the dynamics of market income. A possible explanation is that self-employed children may not yet have started—or may have just started—their own business when they are in their early thirties. Alternatively, or perhaps in addition to this possibility, asset income may grow faster over time for children from more affluent backgrounds. As a result, the bias arising from lifecycle variation can be more pronounced for market income than for earnings. When total income is considered, which also includes government transfers (except taxes), measuring the extent of intergenerational transmission of economic well-being in Canada becomes one step closer to reality. Individuals, both fathers and children, who are less attached to the labour market are now included in the analysis, as long as they have received transfers from governments at any level (i.e., local or national). The literature has shown a strong intergenerational correlation with the receipt of government assistance (Corak, Gustaffsson and Österberg 2004; Page 2004). If children of low-income fathers are more likely to receive government assistance, this may be reflected in the IGE estimates. However, the finding for total income of Table 1 indicates that the intergenerational persistence is only marginally higher, now being close to 0.36. To sum up, the findings presented in Table 1 show that measuring earnings of both fathers and sons in a way that minimizes the impact of transitory earnings and lifecycle bias leads to an estimated IGE for Canada of about 0.32. The elasticity is even higher when market and total income are considered. It is important to emphasize that, while this new finding would put Canada in the middle of the international spectrum of IGE estimates (Björklund and Jäntti 2010; Corak 2013; Solon 2002), very few countries have data that allow an analysis similar to the one presented here. Probably the estimates most comparable to the coefficients presented in this paper are the results in Mazumder (2005). He reported an estimated IGE just above 0.6 in the United States when applying sample restrictions similar to the ones used here. Compared to Mazumder's results, the estimates in this study confirm that Canada remains significantly more mobile than the United States. However, the "true" rate of intergenerational transmission may not be as low as previously thought. ### 4.3 Gender differences in intergenerational income mobility Table 2 replicates the analysis using father–daughter pairs. In general, the intergenerational transmission of earnings and income is weaker for daughters than for sons. The estimated elasticity is about 0.23 for earnings and between 0.24 and 0.25 for income. This result seems to suggest that daughters' outcomes are less dependent on the earnings and income of their fathers. The results from the baseline scenario (scenario 1) can be compared with those of early Canadian studies (Fortin and Lefebvre 1998; Corak 2001), which also found the intergenerational elasticities of earnings and income for fathers and daughters to be about 0.2. Interestingly, unlike the father–son IGE, the father–daughter IGE does not seem to be affected by lifecycle variation. In fact, estimates based on daughters' earnings at age 30 (scenario 1) are very similar to those based on daughters' earnings at age 40 (scenario 2). While this may appear puzzling at first, several factors could help explain this result. Typically, women are more likely than men to experience career breaks related to child-bearing and child rearing during the early stages of their working life. More generally, women are less attached to the labour market than men. The findings are also consistent with results from other countries regarding the role of "assortative mating," which has been identified in the literature as one of the possible reasons for lower IGEs for daughters (see Chadwick and Solon 2002; Ermisch, Francesconi and Siedler 2005; and Raaum et al. 2007). In the presence of marital sorting, daughters with high earnings potential are more likely to marry high-earnings husbands. Such daughters could choose to work fewer hours or accept lower pay in exchange for better work–family balance. In the presence of assortative mating, fathers' lifetime earnings may be more closely tied to the daughters' family (including spousal) earnings than to their own earnings. This could help explain why the estimated father–daughter IGE does not rise as much as it does for sons, when earnings are measured in their forties. Although the estimates in scenario 1 may be biased downward as a result of lifecycle variation, they may be less affected by assortative mating as many daughters remain as single at age 30. On the other hand, at age 40 many daughters are already married. Consequently, low estimated coefficients in scenario 2 point to the possibility that assortative mating may be playing a role in the intergenerational transmission of income between fathers and daughters in Canada. They also suggest that future studies that focus on the extent of intergenerational transmission of income for daughters may need to consider family income. Table 2 Estimates of intergenerational earnings and income elasticity, fathers and daughters Table summary This table displays the results of Estimates of intergenerational earnings and income elasticity. The information is grouped by Proxies for lifetime incomes (appearing as row headers), $\beta$ (fathers and daughters), Earnings, Market income and Total income, calculated using coefficient and standard error units of measure (appearing as column headers). Proxies for lifetime incomes $\beta$ (fathers and daughters) EarningsTable 2 - Note 1 Market incomeTable 2 - Note 2 Total incomeTable 2 - Note 3 coefficient standard error coefficient standard error coefficient standard error Scenario 1 Daughters: at age 30 Fathers: mean over 1978 to 1982 0.191 0.004 0.212 0.003 0.213 0.003 Scenario 2 Daughters: at age 40 Fathers: mean over 1978 to 1982 0.195 0.004 0.207 0.004 0.222 0.003 Scenario 3 Daughters: at age 40 Fathers: mean over ages 35 to 55Table 2 - Note 4 0.221 0.005 0.234 0.005 0.249 0.004 Scenario 4, preferred model Daughters: mean over ages 38 to 42Table 2 - Note 5 Fathers: mean over ages 35 to 55Table 2 - Note 4 0.228 0.004 0.241 0.004 0.254 0.003 ## 5 Nonlinearities The results from the linear regression analysis above may mask nonlinear patterns in the intergenerational transmission of earnings. Various hypotheses have been put forward to explain nonlinearities in the relationship between fathers' and sons' log earnings. For instance, the human capital model proposed by Becker and Tomes (1986) implies a concave pattern of intergenerational earnings transmission, as families at the bottom of the earnings distribution may be more likely to be borrowing-constrained. Empirical evidence related to this hypothesis, however, is rather mixed. In fact, studies of the Nordic European countries find instead a pattern of convexity. Bratsberg et al. (2007), for instance, showed that intergenerational earnings persistence in the Nordic countries is highly nonlinear, with greater mobility found at the bottom of the distribution. Similarly, Björklund, Roine and Waldenström (2012) reported a very high intergenerational persistence of top incomes in Sweden with an estimated elasticity of about 0.9 compared to an average value of 0.26. Subsection 5.1 presents results related to nonlinearities in intergenerational income persistence in Canada. The analysis in the previous section indicates that the channels of intergenerational income transmission seem to be more complicated for daughters than for sons, and may require taking spousal income into account. Therefore, the analysis below is restricted to the father–son sample only. With nearly 200,000 father–son pairs in the IID, it is possible to examine potential nonlinearities at an even finer level of detail than has been done in previous Canadian studies. ### 5.1 Descriptive correlation A descriptive and intuitive method well-suited to illustrating the relationship between fathers' and sons' earnings across the entire spectrum is used to examine the issue of nonlinearity. Fathers are ranked and divided into percentiles according to their lifetime earnings. This generates 100 data points for scatter plots. For each percentile of fathers' earnings, Chart 5 shows the mean log earnings (Panel A) and the earnings shares of sons and fathers (Panel B). The pattern in Panel A clearly shows that the relationship between fathers' and sons' earnings in Canada is not linear. In particular, it reveals a somewhat convex pattern. The estimated slope coefficient is 0.36, but the profile is flat at the bottom of the distribution, increases monotonically over the main part of the distribution and becomes steeper at the top. This pattern has some interesting implications. First, earnings mobility seems to be high among children born to the lowest 15 percentiles of the fathers' earnings distribution, since many of them find themselves in higher earnings percentiles than their fathers. Second, earnings persistence tends to be very high among sons born to the top 10% of the fathers' distribution, suggesting a significant degree of intergenerational transmission of advantages at the top. Third, the intergenerational earnings persistence is moderate between the 15th and 90th percentiles of the fathers' distribution. The mean earnings in a percentile indicate little about earnings variation within the percentile. Panel (B) presents earnings shares held by sons and fathers in each percentile of the fathers' earnings distribution. If sons' lifetime earnings are independent of their fathers', one would expect earnings shares held by sons in each of the fathers' percentiles to be close to 1%. Conversely, a profile that looks like a 45-degree straight line would indicate a high degree of persistence between fathers' and sons' earnings. Panel B generally confirms the nonlinear pattern of intergenerational transmission of earnings noted earlier. Again, mobility appears to be substantial at the bottom of the distribution. For instance, the earnings share held by fathers in the lowest 1% of the distribution is only 0.1%, while the earnings share of sons from these families amount to nearly 0.8% of total sons' earnings. The correlation is stronger in the middle part of the distribution, since both sons' and fathers' earnings shares tend to move in the same direction. Data table for group Chart 5 Data table for group Chart 5 The relationship between fathers' and sons' earnings Table summary This table displays the data for group Chart 5. The information is grouped by Percentile distribution of father's earnings (appearing as row headers), Mean earnings, Earnings share, Fathers and Sons, calculated using log and share units of measure (appearing as column headers). Percentile distribution of father's earnings Mean earnings Earnings share Fathers Sons Fathers Sons log share 1 8.757 10.663 0.0012 0.0074 2 9.315 10.593 0.0021 0.0068 3 9.537 10.699 0.0026 0.0077 4 9.681 10.685 0.0030 0.0075 5 9.785 10.753 0.0033 0.0078 6 9.867 10.707 0.0036 0.0078 7 9.934 10.724 0.0039 0.0077 8 9.994 10.740 0.0041 0.0077 9 10.046 10.735 0.0043 0.0078 10 10.090 10.868 0.0045 0.0093 11 10.131 10.732 0.0047 0.0078 12 10.168 10.734 0.0049 0.0078 13 10.201 10.775 0.0051 0.0082 14 10.232 10.787 0.0052 0.0083 15 10.260 10.746 0.0054 0.0079 16 10.286 10.741 0.0055 0.0079 17 10.310 10.711 0.0057 0.0077 18 10.334 10.761 0.0058 0.0080 19 10.356 10.760 0.0059 0.0085 20 10.376 10.783 0.0060 0.0082 21 10.396 10.788 0.0062 0.0083 22 10.414 10.767 0.0063 0.0082 23 10.431 10.837 0.0064 0.0088 24 10.448 10.825 0.0065 0.0086 25 10.465 10.790 0.0066 0.0084 26 10.480 10.803 0.0067 0.0083 27 10.496 10.816 0.0068 0.0086 28 10.511 10.806 0.0069 0.0084 29 10.525 10.821 0.0070 0.0086 30 10.540 10.835 0.0071 0.0087 31 10.554 10.807 0.0072 0.0085 32 10.567 10.882 0.0073 0.0092 33 10.581 10.840 0.0074 0.0087 34 10.593 10.842 0.0075 0.0087 35 10.606 10.855 0.0076 0.0088 36 10.618 10.901 0.0077 0.0093 37 10.630 10.840 0.0078 0.0087 38 10.642 10.878 0.0079 0.0089 39 10.654 10.852 0.0080 0.0088 40 10.666 10.854 0.0081 0.0089 41 10.677 10.888 0.0082 0.0091 42 10.688 10.925 0.0083 0.0096 43 10.698 10.870 0.0083 0.0090 44 10.709 10.880 0.0084 0.0091 45 10.719 10.900 0.0085 0.0092 46 10.730 10.874 0.0086 0.0090 47 10.740 10.890 0.0087 0.0092 48 10.751 10.978 0.0088 0.0099 49 10.761 10.903 0.0089 0.0092 50 10.771 10.934 0.0090 0.0095 51 10.782 10.910 0.0091 0.0094 52 10.792 10.980 0.0092 0.0102 53 10.803 10.925 0.0092 0.0095 54 10.813 10.918 0.0094 0.0094 55 10.823 10.933 0.0095 0.0095 56 10.834 10.945 0.0095 0.0097 57 10.844 10.933 0.0096 0.0096 58 10.855 10.937 0.0097 0.0096 59 10.865 10.944 0.0099 0.0097 60 10.876 10.963 0.0100 0.0099 61 10.886 10.966 0.0101 0.0096 62 10.897 10.958 0.0102 0.0099 63 10.907 10.962 0.0103 0.0098 64 10.918 10.954 0.0104 0.0098 65 10.929 11.001 0.0105 0.0103 66 10.939 10.991 0.0106 0.0101 67 10.950 10.969 0.0107 0.0099 68 10.961 11.015 0.0108 0.0102 69 10.972 10.977 0.0110 0.0103 70 10.983 10.993 0.0111 0.0101 71 10.994 11.058 0.0112 0.0108 72 11.006 10.997 0.0113 0.0101 73 11.018 11.020 0.0115 0.0104 74 11.030 11.024 0.0116 0.0106 75 11.042 11.017 0.0118 0.0105 76 11.054 11.044 0.0119 0.0108 77 11.067 11.054 0.0121 0.0109 78 11.081 11.057 0.0122 0.0107 79 11.094 11.112 0.0124 0.0115 80 11.108 11.070 0.0126 0.0108 81 11.123 11.116 0.0127 0.0116 82 11.138 11.091 0.0129 0.0112 83 11.153 11.106 0.0131 0.0115 84 11.169 11.130 0.0134 0.0117 85 11.186 11.124 0.0136 0.0115 86 11.203 11.147 0.0138 0.0118 87 11.221 11.157 0.0141 0.0121 88 11.241 11.151 0.0143 0.0120 89 11.262 11.148 0.0146 0.0121 90 11.285 11.156 0.0150 0.0118 91 11.311 11.153 0.0154 0.0119 92 11.339 11.212 0.0158 0.0126 93 11.371 11.254 0.0163 0.0132 94 11.410 11.260 0.0170 0.0131 95 11.455 11.273 0.0178 0.0134 96 11.508 11.365 0.0187 0.0146 97 11.578 11.377 0.0201 0.0153 98 11.682 11.472 0.0223 0.0164 99 11.868 11.490 0.0269 0.0166 100 12.634 11.979 0.0573 0.0280 The patterns of nonlinearities observed in Chart 5 do not appear consistent with the borrowing constraints model in Becker and Tomes (1986) or Mulligan (1997), which would imply a concave relationship between children's and parents' earnings. These studies suggest that parents in the lower half of the distribution are more likely to be credit-constrained. Under some additional assumptions, this would imply sub-optimal investments in children's human capital at the bottom of the fathers' earnings distribution, and thus a resulting stronger correlation (i.e., steeper profile) between father and son earnings. Credit constraints are then assumed to gradually relax at higher percentiles of the distribution; as a result, sons' earnings would become more independent from fathers' earnings in the upper half of the distribution. A possible reason why a pattern of concavity may not be found in the data is the violation of some of the model's assumptions. Han and Mulligan (2001), for instance, showed that heterogeneity in children's innate earnings potential, as well as in parents' altruism, make testing for the existence of credit-constrained families difficult. The pattern of nonlinearity observed in the Canadian data, however, seems to be more in line with the Nordic evidence of a convex intergenerational earnings relationship. Bratsberg et al. (2007) argued that institutional factors may explain why mobility is higher at the bottom. In particular, they suggested that educational and welfare systems in Nordic countries help the upward mobility of young people with few parental resources. Interestingly, the findings of this study for Canada are different from the patterns estimated in the United States, which exhibit an almost perfectly linear relationship between children's and parents' ranks in the income distribution (Chetty et al. 2014).Note 14 In sum, the nonlinear patterns in Chart 5 suggest that a single IGE coefficient may be insufficient to offer an accurate picture of intergenerational earnings mobility in Canada. Corak and Heisz (1999) reached a similar conclusion using income transition matrices and then proceeded to employ a nonparametric technique to explore the nature of these nonlinearities. In what follows, two different parametric approaches are used instead—that is, higher-order polynomial and quantile regressions—to address the issue of nonlinearities. The parametric results from this study will be briefly compared to the findings of Corak and Heisz (1999) in the last section of this paper. ### 5.2 Nonlinear regressions: higher-order polynomial As shown by Bratsberg et al. (2007), more flexible functional forms (i.e., higher-order polynomial) can be used to estimate the intergenerational earnings model in order to fit the data. Here, the coefficients from a fourth-order polynomial are estimated on the IID data and used to calculate the IGE at each percentile of the fathers' earnings distribution. The results, presented in Chart 6, confirm that the degree of intergenerational earnings mobility in Canada is characterized by a marked nonlinear pattern. In fact, Chart 6 shows a logarithmic growth trajectory. In particular, the elasticity is quite low for sons at the bottom two percentiles of the fathers' earnings distribution (about 0.1); this suggests a significant degree of upward mobility for sons born to very low-earnings fathers. Further up the fathers' earnings distribution, the degree of mobility starts to decline. The estimated IGE sharply rises, reaching 0.32—the same value as the estimated coefficient from the linear specification—at about the 23rd percentile. Thereafter, earnings persistence increases monotonically with fathers' earnings. The estimated elasticity tops at 0.45 for the 98th and 99th percentiles of fathers' earnings, suggesting a strong intergenerational correlation of earnings among very high-earnings families. A salient feature of Chart 6 is that the estimates from the linear model (0.32) seem to understate the intergenerational transmission of earnings for the bulk of the population. These results support the conclusions in Bratsberg et al. (2007) that international comparisons of intergenerational earnings mobility based on more flexible specifications may be more meaningful. Indeed, comparing the results in this study with the findings in Bratsberg et al. suggests that Canada has a pattern of intergenerational earnings mobility that is quite similar to the one observed in the Scandinavian countries: a flat intergenerational relationship in the lower segments of the fathers' distribution and an increasingly positive correlation in middle and upper segments. Interestingly, this pattern is not observed in the United States or in the United Kingdom (Bratsberg et al. 2007). Data table for Chart 6 Data table for Chart 6 Fourth-order polynomial estimate of $\beta$ (IGE), evaluated at each percentile of fathers' earnings, father-son pairs Table summary This table displays the data for Chart 6. The information is grouped by Percentile of fathers' lifetime earnings (appearing as row headers), Fourth-order polynomial (fathers' earnings), 95% confidence interval (lower), 95% confidence interval (upper) and $\beta$=0.318 in linear model, calculated using $\beta$ (IGE) units of measure (appearing as column headers). Percentile of fathers' lifetime earnings Fourth-order polynomial (fathers' earnings) 95% confidence interval (lower) 95% confidence interval (upper) $\beta$ = 0.318 in linear model $\beta$ (IGE) 0 Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable 1 0.0743 0.0528 0.0958 0.318 2 0.1367 0.1189 0.1546 0.318 3 0.1719 0.1560 0.1879 0.318 4 0.1958 0.1812 0.2104 0.318 5 0.2134 0.1998 0.2270 0.318 6 0.2274 0.2146 0.2402 0.318 7 0.2394 0.2273 0.2514 0.318 8 0.2496 0.2382 0.2611 0.318 9 0.2584 0.2475 0.2694 0.318 10 0.2661 0.2556 0.2766 0.318 11 0.2731 0.2629 0.2832 0.318 12 0.2790 0.2693 0.2888 0.318 13 0.2846 0.2751 0.2940 0.318 14 0.2897 0.2805 0.2989 0.318 15 0.2942 0.2852 0.3032 0.318 16 0.2984 0.2896 0.3072 0.318 17 0.3025 0.2939 0.3110 0.318 18 0.3062 0.2978 0.3146 0.318 19 0.3098 0.3015 0.3180 0.318 20 0.3130 0.3049 0.3212 0.318 21 0.3161 0.3081 0.3241 0.318 22 0.3189 0.3109 0.3268 0.318 23 0.3217 0.3138 0.3295 0.318 24 0.3243 0.3165 0.3321 0.318 25 0.3268 0.3191 0.3345 0.318 26 0.3293 0.3216 0.3369 0.318 27 0.3316 0.3240 0.3392 0.318 28 0.3338 0.3262 0.3414 0.318 29 0.3360 0.3285 0.3436 0.318 30 0.3381 0.3306 0.3457 0.318 31 0.3402 0.3327 0.3477 0.318 32 0.3422 0.3347 0.3497 0.318 33 0.3441 0.3366 0.3516 0.318 34 0.3460 0.3385 0.3535 0.318 35 0.3478 0.3403 0.3553 0.318 36 0.3496 0.3421 0.3571 0.318 37 0.3512 0.3437 0.3588 0.318 38 0.3529 0.3454 0.3605 0.318 39 0.3546 0.3470 0.3622 0.318 40 0.3562 0.3486 0.3638 0.318 41 0.3578 0.3501 0.3654 0.318 42 0.3592 0.3516 0.3669 0.318 43 0.3607 0.3530 0.3684 0.318 44 0.3622 0.3544 0.3699 0.318 45 0.3636 0.3558 0.3714 0.318 46 0.3650 0.3572 0.3728 0.318 47 0.3664 0.3585 0.3743 0.318 48 0.3678 0.3598 0.3757 0.318 49 0.3691 0.3611 0.3771 0.318 50 0.3705 0.3624 0.3785 0.318 51 0.3719 0.3638 0.3800 0.318 52 0.3732 0.3650 0.3813 0.318 53 0.3745 0.3663 0.3827 0.318 54 0.3758 0.3675 0.3841 0.318 55 0.3771 0.3688 0.3855 0.318 56 0.3784 0.3700 0.3869 0.318 57 0.3797 0.3712 0.3882 0.318 58 0.3810 0.3724 0.3896 0.318 59 0.3823 0.3737 0.3910 0.318 60 0.3836 0.3748 0.3923 0.318 61 0.3849 0.3761 0.3937 0.318 62 0.3861 0.3772 0.3950 0.318 63 0.3874 0.3784 0.3963 0.318 64 0.3886 0.3796 0.3977 0.318 65 0.3899 0.3807 0.3990 0.318 66 0.3911 0.3819 0.4004 0.318 67 0.3923 0.3830 0.4017 0.318 68 0.3936 0.3841 0.4030 0.318 69 0.3948 0.3853 0.4043 0.318 70 0.3961 0.3864 0.4057 0.318 71 0.3973 0.3875 0.4071 0.318 72 0.3986 0.3887 0.4084 0.318 73 0.3999 0.3899 0.4098 0.318 74 0.4012 0.3911 0.4113 0.318 75 0.4025 0.3923 0.4127 0.318 76 0.4038 0.3934 0.4141 0.318 77 0.4051 0.3947 0.4156 0.318 78 0.4065 0.3959 0.4171 0.318 79 0.4079 0.3971 0.4186 0.318 80 0.4093 0.3984 0.4202 0.318 81 0.4107 0.3997 0.4218 0.318 82 0.4122 0.4009 0.4234 0.318 83 0.4136 0.4022 0.4250 0.318 84 0.4151 0.4036 0.4267 0.318 85 0.4166 0.4049 0.4284 0.318 86 0.4181 0.4062 0.4301 0.318 87 0.4198 0.4076 0.4319 0.318 88 0.4215 0.4091 0.4339 0.318 89 0.4233 0.4106 0.4359 0.318 90 0.4252 0.4123 0.4382 0.318 91 0.4273 0.4140 0.4405 0.318 92 0.4294 0.4158 0.4430 0.318 93 0.4318 0.4177 0.4458 0.318 94 0.4344 0.4198 0.4490 0.318 95 0.4373 0.4220 0.4525 0.318 96 0.4403 0.4243 0.4564 0.318 97 0.4438 0.4265 0.4611 0.318 98 0.4475 0.4280 0.4670 0.318 99 0.4475 0.4222 0.4729 0.318 100 Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable ### 5.3 Quantile regressions While the results above present evidence of nonlinearity in the IGE across the distribution of fathers' earnings, it is also interesting to examine whether mobility differs across the distribution of sons' earnings. As pointed out in Mulligan (1997) and Corak and Heisz (1999), the optimal amount of human capital investment made by fathers may also depend—positively—upon the child's ability. These authors note that the parents who are more likely to be credit-constrained are low-income parents with high-ability children. This raises the question of how the earnings of high-ability children from low-income backgrounds compare to those of their counterparts from high-income families. The answer to this question may be informative for policy discussions related to equality of opportunity in Canada. This notion refers to the existence of similar economic opportunities for children with similar abilities, regardless of their family of origin (Roemer 1998). Grawe (2004) suggested using quantile regressions to investigate this type of question. For instance, high elasticity in the top percentiles of children's earnings distribution would suggest that high-earnings children come almost exclusively from families with high-earnings fathers. Unless one makes particular assumptions regarding the heritability of innate ability, this can be interpreted as evidence for inequality of opportunity. That is, it would suggest that high-ability children from low-earnings backgrounds have little chances to realize their full potential and become high-income earners. Following the approach in Grawe (2004), Chart 7 presents the results from quantile regressions estimated on the father–son pairs used in this study. The estimates are produced for each percentile (1st, 2nd…, 99th) of sons' earnings. The estimated 95% confidence intervals are also reported. The quantile regression results depict a clear pattern of nonlinearity over the distribution of sons' earnings. The estimated earnings elasticity is relatively low (about 0.2) at the bottom two percentiles of the sons' earnings distribution, rises to 0.35 at the 10th percentile, and stays at about 0.33 to 0.34 throughout the lower half of the distribution. It then drops gradually, reaching 0.27 at the 80th to the 85th percentiles, and reverses its course by rising again to 0.41 at the 99th percentile. Three distinct patterns of intergenerational earnings transmission emerge from the quantile regression analysis. First, quantile regression coefficients are quite low for sons in the bottom two percentiles of the distribution. That is, a large fraction of sons with the lowest earnings do not have fathers who were themselves at the very bottom of the distribution in their generation. This implies that the sons of the lowest-earnings fathers are relatively mobile, which is consistent with the results from the polynomial model above. Second, from the 10th to the 85th percentiles of the distribution of the sons' earnings, persistence is fairly high in the beginning (the estimated elasticity is above 0.35 at the 10th percentile), declining slowly over the lower half, and dropping more notably over the upper-middle part of the distribution. A rather high elasticity over the lower-middle part of the sons' distribution indicates that a considerable fraction of moderate-earnings sons have fathers with similar economic outcomes. Such correlation, however, becomes less obvious as sons' earnings increase. Earnings mobility starts to rise for sons in the upper-middle part of the distribution. The estimated elasticity declines from 0.33 at the median to 0.27 at the 85th percentile. This suggests that sons from a wide range of earnings backgrounds have a good chance of becoming adults with above-average earnings. Third, Chart 7 shows the reversal of the quantile regression coefficients over the top (85th to 99th) percentiles, which reveals a stronger degree of earnings persistence for high-earnings sons. The estimated coefficients are 0.38 and 0.41 for the top two percentiles, respectively. That is, a significant fraction of those who make it to the highest-earnings groups have a high-earnings father. Finally, Chart 7 also suggests that Canadian earnings mobility may be characterized by rather complex transmission mechanisms. In particular, distinct channels of parental influence may be at work at different parts of the distribution. At the bottom of the distribution, institutional factors—as suggested in the Nordic studies—may play a role in facilitating upward mobility for children from very low-earnings backgrounds.Note 15 The estimated profile over the bulk of the distribution would be broadly in line with the credit constraints hypothesis: the relationship between moderate-earnings sons and moderate-earnings fathers is expected to be stronger because these fathers are more likely to be credit-constrained and may not be able to invest optimally in the human capital of their children. The profile corresponding to the upper-middle part of the distribution may also be consistent with less binding credit constraints for fathers whose earnings are increasingly sufficient to invest in their children's human capital. Finally, channels of transmission may be even more complex at the top, as suggested in the literature. Factors other than human capital investment, such as networking or family-specific capital, may be more salient for the intergenerational earnings transmission among top-earnings families (Björklund, Roine and Waldenström 2012; Corak and Piraino 2011; Kramarz and Skans 2014). A few caveats, however, need to be borne in mind in interpreting the dynamics at the bottom end of the distribution. It is possible that, at this end of the distribution, the mother is carrying the load in terms of market activities. Also note that the analytical sample used in this study excludes all those sons raised in fatherless (i.e., single-mother) households who are more likely to be in low-income situations. Moreover, data quality may also be a concern, since measurement error often tends to be more pronounced at the bottom end of earnings or income distribution. Further discussion on and identification of channels of influence is required, but is beyond the scope of this paper. Data table for Chart 7 Chart 7 Quantile regression estimates of $\beta$ (IGE), earnings of father–son pairs Table summary This table displays the data for Chart 7. The information is grouped by Percentile of sons' earnings (appearing as row headers), Estimate of $\beta$ (IGE), 95% confidence interval (lower) and 95% confidence interval (upper), calculated using $\beta$ (IGE) units of measure (appearing as column headers). Percentile of sons' earnings Estimate of $\beta$ (IGE) 95% confidence interval (lower) 95% confidence interval (upper) $\beta$ (IGE) 0 Note ...: not applicable Note ...: not applicable Note ...: not applicable 1 0.1969 0.1415 0.2523 2 0.2519 0.2188 0.2850 3 0.2915 0.2649 0.3181 4 0.3073 0.2860 0.3287 5 0.3290 0.3083 0.3497 6 0.3370 0.3175 0.3564 7 0.3479 0.3315 0.3643 8 0.3501 0.3350 0.3653 9 0.3524 0.3385 0.3663 10 0.3528 0.3401 0.3656 11 0.3532 0.3401 0.3663 12 0.3537 0.3416 0.3659 13 0.3543 0.3434 0.3652 14 0.3561 0.3451 0.3671 15 0.3516 0.3412 0.3620 16 0.3505 0.3408 0.3602 17 0.3496 0.3400 0.3592 18 0.3489 0.3404 0.3575 19 0.3461 0.3372 0.3551 20 0.3459 0.3372 0.3546 21 0.3450 0.3367 0.3534 22 0.3454 0.3376 0.3532 23 0.3439 0.3361 0.3517 24 0.3431 0.3358 0.3504 25 0.3432 0.3360 0.3503 26 0.3421 0.3350 0.3492 27 0.3408 0.3343 0.3473 28 0.3401 0.3335 0.3467 29 0.3394 0.3332 0.3456 30 0.3392 0.3329 0.3456 31 0.3390 0.3328 0.3453 32 0.3388 0.3326 0.3450 33 0.3394 0.3334 0.3453 34 0.3393 0.3332 0.3455 35 0.3382 0.3324 0.3440 36 0.3372 0.3312 0.3432 37 0.3373 0.3315 0.3431 38 0.3366 0.3309 0.3423 39 0.3359 0.3303 0.3415 40 0.3350 0.3293 0.3406 41 0.3334 0.3280 0.3387 42 0.3321 0.3266 0.3376 43 0.3308 0.3256 0.3359 44 0.3302 0.3250 0.3354 45 0.3294 0.3241 0.3347 46 0.3285 0.3231 0.3338 47 0.3278 0.3224 0.3332 48 0.3268 0.3216 0.3321 49 0.3257 0.3202 0.3312 50 0.3237 0.3184 0.3290 51 0.3221 0.3165 0.3277 52 0.3187 0.3134 0.3240 53 0.3170 0.3121 0.3219 54 0.3155 0.3106 0.3204 55 0.3138 0.3084 0.3192 56 0.3120 0.3071 0.3170 57 0.3112 0.3061 0.3164 58 0.3096 0.3046 0.3147 59 0.3080 0.3027 0.3133 60 0.3065 0.3011 0.3119 61 0.3048 0.2997 0.3100 62 0.3040 0.2988 0.3092 63 0.3025 0.2972 0.3079 64 0.3002 0.2945 0.3058 65 0.2978 0.2925 0.3030 66 0.2959 0.2904 0.3014 67 0.2941 0.2889 0.2993 68 0.2923 0.2872 0.2975 69 0.2897 0.2843 0.2951 70 0.2877 0.2821 0.2933 71 0.2859 0.2805 0.2914 72 0.2840 0.2784 0.2896 73 0.2837 0.2781 0.2893 74 0.2822 0.2767 0.2877 75 0.2800 0.2742 0.2858 76 0.2795 0.2732 0.2858 77 0.2780 0.2719 0.2841 78 0.2771 0.2712 0.2829 79 0.2746 0.2681 0.2812 80 0.2732 0.2665 0.2799 81 0.2720 0.2653 0.2788 82 0.2725 0.2653 0.2798 83 0.2723 0.2655 0.2791 84 0.2738 0.2661 0.2815 85 0.2742 0.2666 0.2817 86 0.2749 0.2665 0.2833 87 0.2774 0.2684 0.2864 88 0.2793 0.2696 0.2890 89 0.2819 0.2717 0.2921 90 0.2847 0.2739 0.2956 91 0.2926 0.2804 0.3048 92 0.2956 0.2828 0.3084 93 0.3006 0.2859 0.3153 94 0.3101 0.2932 0.3270 95 0.3179 0.2975 0.3382 96 0.3301 0.3054 0.3549 97 0.3406 0.3096 0.3717 98 0.3683 0.3246 0.4120 99 0.4131 0.3382 0.4879 100 Note ...: not applicable Note ...: not applicable Note ...: not applicable ### 5.4 Market income and total income Chart 8 shows nonlinear patterns in intergenerational income persistence for market (dashed line) and total (dotted line) income. Both polynomial (Panel A) and quantile (Panel B) regression results are presented. In general, despite differences in magnitude, the pattern of nonlinearity across the distribution seems quite similar for all three income measures. As in the linear model, persistence tends to be higher for market or total income than for earnings. This is especially the case at the very top: sons of fathers with the highest income are much more likely to have the highest incomes as adults. Differences in generational persistence for different income sources, however, are more visible in the quantile regression results (Panel B). For instance, the correlation between low-income sons (e.g., those in the 10th percentile) and low-income fathers is stronger than it is between low-earnings fathers and sons. Moreover, higher mobility over the upper-middle part of the distribution is somewhat limited for market and, especially, total income. Unlike the results for earnings, where some of the higher-earnings sons (e.g., those at the 85th percentile) can come from a rather diverse earnings background, the higher-income sons are more likely to have fathers with a similar position in the income distribution. Income persistence is especially pronounced at the top end of the distribution, at 0.51 or higher for the top two percentiles of the sons' total income. These results are broadly consistent with a growing literature showing that affluent families are able to improve the income potential of their children in various ways—including better child care, private schooling, transfer of entrepreneurial skills and social connections (e.g., Duncan and Murnane 2011). Data table for group Chart 8 Data table for group Chart 8 Nonlinear estimates of β (intergenerational income elasticity) for fathers and sons, by income source Table summary This table displays the results of Nonlinear estimates of β (intergenerational income elasticity) for fathers and sons. The information is grouped by Percentile (appearing as row headers), Polynomial regressions, Quantile regressions, Earnings, Market income and Total income, calculated using estimate units of measure (appearing as column headers). Percentile Polynomial regressionsNote 1 Quantile regressionsNote 2 Earnings Market income Total income Earnings Market income Total income estimate 1 0.0743 0.0940 0.1152 0.1969 0.2301 0.2023 2 0.1367 0.1421 0.1601 0.2519 0.2666 0.2318 3 0.1719 0.1713 0.1864 0.2915 0.2922 0.2611 4 0.1958 0.1911 0.2057 0.3073 0.3150 0.2989 5 0.2134 0.2067 0.2209 0.3290 0.3278 0.3331 6 0.2274 0.2201 0.2333 0.3370 0.3412 0.3530 7 0.2394 0.2312 0.2438 0.3479 0.3469 0.3633 8 0.2496 0.2410 0.2531 0.3501 0.3573 0.3713 9 0.2584 0.2497 0.2613 0.3524 0.3612 0.3779 10 0.2661 0.2576 0.2689 0.3528 0.3624 0.3780 11 0.2731 0.2649 0.2755 0.3532 0.3645 0.3772 12 0.2790 0.2714 0.2818 0.3537 0.3660 0.3756 13 0.2846 0.2776 0.2876 0.3543 0.3662 0.3749 14 0.2897 0.2833 0.2931 0.3561 0.3667 0.3741 15 0.2942 0.2885 0.2981 0.3516 0.3672 0.3702 16 0.2984 0.2934 0.3028 0.3505 0.3668 0.3689 17 0.3025 0.2979 0.3072 0.3496 0.3660 0.3651 18 0.3062 0.3022 0.3115 0.3489 0.3633 0.3626 19 0.3098 0.3062 0.3154 0.3461 0.3634 0.3623 20 0.3130 0.3101 0.3190 0.3459 0.3613 0.3605 21 0.3161 0.3137 0.3225 0.3450 0.3585 0.3583 22 0.3189 0.3171 0.3259 0.3454 0.3565 0.3584 23 0.3217 0.3203 0.3291 0.3439 0.3551 0.3575 24 0.3243 0.3234 0.3321 0.3431 0.3549 0.3580 25 0.3268 0.3264 0.3350 0.3432 0.3521 0.3565 26 0.3293 0.3292 0.3379 0.3421 0.3516 0.3551 27 0.3316 0.3320 0.3406 0.3408 0.3493 0.3558 28 0.3338 0.3346 0.3431 0.3401 0.3486 0.3549 29 0.3360 0.3371 0.3456 0.3394 0.3467 0.3543 30 0.3381 0.3394 0.3480 0.3392 0.3458 0.3543 31 0.3402 0.3417 0.3503 0.3390 0.3449 0.3531 32 0.3422 0.3439 0.3526 0.3388 0.3439 0.3525 33 0.3441 0.3461 0.3548 0.3394 0.3436 0.3521 34 0.3460 0.3482 0.3569 0.3393 0.3426 0.3519 35 0.3478 0.3502 0.3589 0.3382 0.3424 0.3521 36 0.3496 0.3522 0.3610 0.3372 0.3416 0.3515 37 0.3512 0.3542 0.3630 0.3373 0.3407 0.3513 38 0.3529 0.3560 0.3649 0.3366 0.3398 0.3502 39 0.3546 0.3579 0.3668 0.3359 0.3389 0.3501 40 0.3562 0.3596 0.3686 0.3350 0.3382 0.3490 41 0.3578 0.3614 0.3704 0.3334 0.3369 0.3484 42 0.3592 0.3631 0.3722 0.3321 0.3358 0.3467 43 0.3607 0.3648 0.3738 0.3308 0.3349 0.3469 44 0.3622 0.3664 0.3755 0.3302 0.3337 0.3462 45 0.3636 0.3680 0.3772 0.3294 0.3327 0.3462 46 0.3650 0.3695 0.3788 0.3285 0.3314 0.3456 47 0.3664 0.3711 0.3803 0.3278 0.3308 0.3445 48 0.3678 0.3726 0.3819 0.3268 0.3284 0.3440 49 0.3691 0.3741 0.3835 0.3257 0.3267 0.3426 50 0.3705 0.3756 0.3850 0.3237 0.3260 0.3415 51 0.3719 0.3770 0.3865 0.3221 0.3244 0.3409 52 0.3732 0.3785 0.3880 0.3187 0.3236 0.3400 53 0.3745 0.3799 0.3895 0.3170 0.3220 0.3387 54 0.3758 0.3813 0.3910 0.3155 0.3201 0.3369 55 0.3771 0.3827 0.3924 0.3138 0.3188 0.3357 56 0.3784 0.3842 0.3938 0.3120 0.3178 0.3352 57 0.3797 0.3855 0.3953 0.3112 0.3169 0.3338 58 0.3810 0.3869 0.3967 0.3096 0.3156 0.3332 59 0.3823 0.3883 0.3981 0.3080 0.3131 0.3314 60 0.3836 0.3897 0.3995 0.3065 0.3123 0.3310 61 0.3849 0.3911 0.4010 0.3048 0.3106 0.3292 62 0.3861 0.3925 0.4024 0.3040 0.3103 0.3276 63 0.3874 0.3938 0.4037 0.3025 0.3099 0.3265 64 0.3886 0.3952 0.4051 0.3002 0.3086 0.3259 65 0.3899 0.3966 0.4065 0.2978 0.3072 0.3248 66 0.3911 0.3979 0.4080 0.2959 0.3056 0.3243 67 0.3923 0.3992 0.4094 0.2941 0.3037 0.3228 68 0.3936 0.4006 0.4108 0.2923 0.3025 0.3220 69 0.3948 0.4019 0.4123 0.2897 0.3014 0.3209 70 0.3961 0.4033 0.4137 0.2877 0.3001 0.3204 71 0.3973 0.4047 0.4151 0.2859 0.3006 0.3201 72 0.3986 0.4061 0.4166 0.2840 0.3002 0.3191 73 0.3999 0.4075 0.4181 0.2837 0.2997 0.3197 74 0.4012 0.4089 0.4196 0.2822 0.2990 0.3189 75 0.4025 0.4104 0.4210 0.2800 0.2988 0.3186 76 0.4038 0.4118 0.4226 0.2795 0.2986 0.3181 77 0.4051 0.4133 0.4241 0.2780 0.2974 0.3189 78 0.4065 0.4148 0.4257 0.2771 0.2960 0.3189 79 0.4079 0.4163 0.4273 0.2746 0.2964 0.3190 80 0.4093 0.4179 0.4289 0.2732 0.2972 0.3194 81 0.4107 0.4195 0.4306 0.2720 0.2966 0.3199 82 0.4122 0.4211 0.4323 0.2725 0.2978 0.3226 83 0.4136 0.4228 0.4340 0.2723 0.3004 0.3250 84 0.4151 0.4245 0.4358 0.2738 0.3033 0.3269 85 0.4166 0.4262 0.4376 0.2742 0.3064 0.3295 86 0.4181 0.4280 0.4396 0.2749 0.3114 0.3336 87 0.4198 0.4299 0.4416 0.2774 0.3144 0.3386 88 0.4215 0.4320 0.4438 0.2793 0.3183 0.3433 89 0.4233 0.4342 0.4460 0.2819 0.3245 0.3484 90 0.4252 0.4364 0.4484 0.2847 0.3318 0.3566 91 0.4273 0.4389 0.4511 0.2926 0.3417 0.3656 92 0.4294 0.4417 0.4540 0.2956 0.3496 0.3754 93 0.4318 0.4447 0.4571 0.3006 0.3617 0.3894 94 0.4344 0.4483 0.4606 0.3101 0.3740 0.4035 95 0.4373 0.4519 0.4644 0.3179 0.3898 0.4235 96 0.4403 0.4563 0.4689 0.3301 0.4124 0.4454 97 0.4438 0.4615 0.4737 0.3406 0.4421 0.4752 98 0.4475 0.4662 0.4777 0.3683 0.4765 0.5096 99 0.4475 0.4671 0.4758 0.4131 0.5359 0.5765 ### 5.5 Lifecycle variation and nonlinearities As it was the case in the simple linear model, the patterns of nonlinearity may not be estimated correctly if the lifecycle and errors-in-variables biases are not properly addressed. The distortion is likely to be greater in the upper part of the distribution, since most children with high lifetime earnings have not yet reached their full earnings potential at younger ages. To confirm this intuition, the analysis in Subsection 5.4 is repeated with two alternative definitions for fathers' and sons' lifetime earnings. The first scenario defines fathers' lifetime earnings as five-year averages over the period from 1978 to 1982, while sons' earnings are drawn from the year corresponding to age 30 (dashed line). The second scenario maintains the same earnings definition for fathers but measures sons' earnings at age 40 (dotted line). As the results in Subsection 5.4 indicate, both lifecycle and errors-in-variables biases are likely to be present in the first scenario, while the former bias can be minimized in the second scenario. The results are presented in Chart 9, which also provides the preferred estimates for reference (solid line). Chart 9, indeed, reveals a very different pattern of nonlinearity when sons' earnings at age 30 are used. For polynomial regressions, this leads to an underestimation of intergenerational persistence throughout virtually the entire distribution of fathers' earnings. Because of lifecycle variation, the extent of the correlation between fathers' and sons' earnings appears to be particularly underestimated in the upper part of the distribution. For instance, the estimated elasticity at the 95th percentile is only 0.24—about 45% lower than the estimate (0.44) from the preferred definition. When lifecycle bias is minimized by using sons' earnings at age 40, the pattern of nonlinearity is almost identical to the pattern yielded by the preferred model for much of the distribution. However, the estimates for the top percentiles are still lower than those in the preferred model because the measure of fathers' earnings is less accurate. The quantile regression results also show that not accounting for lifecycle variation may result in significantly biased patterns of nonlinearity. The decline in earnings persistence from the 15th to the 95th percentiles of sons' earnings is much steeper than that produced by the preferred model and, thus, gives a wrong impression that mobility increases until the 95th percentile of the sons' distribution. The estimated elasticity at the 95th percentile is as low as 0.14, when sons' earnings at age 30 are used as a proxy for lifetime earnings, 56% lower than the preferred estimate of 0.32. Again, the pattern of nonlinearity becomes more similar to the one from the preferred model when sons' earnings at age 40 are used. It is also interesting to note that using five-year earnings averages from a given time period for fathers also leads to a downward bias in the quantile regression estimates for large segments in the middle of the sons' earnings distribution. Data table for group Chart 9 Data table for group Chart 9 Lifecycle variation and nonlinearities Table summary This table displays the results of Lifecycle variation and nonlinearities. The information is grouped by Percentile (appearing as row headers), Polynomial regressions, Quantile regressions, Fathers (mean over age 35 to 55), sons (mean over age 38 to 42), Fathers (mean over year 1978 to 1982), sons (at age 30) and Fathers (mean over year 1978 to 1982), sons (at age 40), calculated using β (intergenerational income elasticity) estimate units of measure (appearing as column headers). Percentile Polynomial regressionsChart 9 - note 1 Quantile regressionsChart 9 - note 2 Fathers (mean over age 35 to 55), sons (mean over age 38 to 42) Fathers (mean over year 1978 to 1982), sons (at age 30) Fathers (mean over year 1978 to 1982), sons (at age 40) Fathers (mean over age 35 to 55), sons (mean over age 38 to 42) Fathers (mean over year 1978 to 1982), sons (at age 30) Fathers (mean over year 1978 to 1982), sons (at age 40) β (IGE) estimate 1 0.0743 0.1201 0.0287 0.1969 0.1229 0.1703 2 0.1367 0.1820 0.1074 0.2519 0.1872 0.2504 3 0.1719 0.2076 0.1504 0.2915 0.2053 0.2960 4 0.1958 0.2223 0.1793 0.3073 0.2278 0.3045 5 0.2134 0.2314 0.1999 0.3290 0.2397 0.3184 6 0.2274 0.2378 0.2161 0.3370 0.2527 0.3352 7 0.2394 0.2427 0.2294 0.3479 0.2653 0.3443 8 0.2496 0.2464 0.2403 0.3501 0.2759 0.3495 9 0.2584 0.2493 0.2497 0.3524 0.2853 0.3564 10 0.2661 0.2516 0.2576 0.3528 0.2900 0.3579 11 0.2731 0.2536 0.2647 0.3532 0.3023 0.3573 12 0.2790 0.2552 0.2709 0.3537 0.3061 0.3575 13 0.2846 0.2565 0.2762 0.3543 0.3137 0.3539 14 0.2897 0.2576 0.2811 0.3561 0.3203 0.3503 15 0.2942 0.2586 0.2857 0.3516 0.3224 0.3436 16 0.2984 0.2594 0.2898 0.3505 0.3186 0.3402 17 0.3025 0.2601 0.2935 0.3496 0.3188 0.3380 18 0.3062 0.2608 0.2972 0.3489 0.3201 0.3336 19 0.3098 0.2614 0.3004 0.3461 0.3187 0.3293 20 0.3130 0.2619 0.3035 0.3459 0.3184 0.3264 21 0.3161 0.2623 0.3064 0.3450 0.3170 0.3258 22 0.3189 0.2627 0.3092 0.3454 0.3151 0.3225 23 0.3217 0.2631 0.3118 0.3439 0.3122 0.3210 24 0.3243 0.2634 0.3144 0.3431 0.3086 0.3199 25 0.3268 0.2637 0.3168 0.3432 0.3060 0.3183 26 0.3293 0.2639 0.3191 0.3421 0.3021 0.3171 27 0.3316 0.2642 0.3213 0.3408 0.2979 0.3144 28 0.3338 0.2643 0.3233 0.3401 0.2947 0.3120 29 0.3360 0.2645 0.3253 0.3394 0.2908 0.3091 30 0.3381 0.2647 0.3273 0.3392 0.2866 0.3080 31 0.3402 0.2648 0.3292 0.3390 0.2839 0.3078 32 0.3422 0.2650 0.3310 0.3388 0.2818 0.3063 33 0.3441 0.2651 0.3328 0.3394 0.2790 0.3057 34 0.3460 0.2652 0.3346 0.3393 0.2760 0.3053 35 0.3478 0.2652 0.3363 0.3382 0.2719 0.3038 36 0.3496 0.2653 0.3379 0.3372 0.2692 0.3029 37 0.3512 0.2653 0.3394 0.3373 0.2649 0.3028 38 0.3529 0.2654 0.3409 0.3366 0.2641 0.3006 39 0.3546 0.2654 0.3424 0.3359 0.2619 0.2990 40 0.3562 0.2654 0.3439 0.3350 0.2596 0.2971 41 0.3578 0.2654 0.3453 0.3334 0.2567 0.2954 42 0.3592 0.2654 0.3467 0.3321 0.2552 0.2940 43 0.3607 0.2654 0.3480 0.3308 0.2517 0.2934 44 0.3622 0.2654 0.3494 0.3302 0.2488 0.2911 45 0.3636 0.2654 0.3507 0.3294 0.2464 0.2888 46 0.3650 0.2653 0.3520 0.3285 0.2450 0.2869 47 0.3664 0.2653 0.3533 0.3278 0.2423 0.2854 48 0.3678 0.2652 0.3546 0.3268 0.2408 0.2841 49 0.3691 0.2652 0.3559 0.3257 0.2386 0.2838 50 0.3705 0.2651 0.3571 0.3237 0.2370 0.2819 51 0.3719 0.2650 0.3583 0.3221 0.2340 0.2797 52 0.3732 0.2649 0.3596 0.3187 0.2323 0.2791 53 0.3745 0.2648 0.3609 0.3170 0.2291 0.2794 54 0.3758 0.2647 0.3621 0.3155 0.2260 0.2781 55 0.3771 0.2646 0.3633 0.3138 0.2234 0.2763 56 0.3784 0.2645 0.3645 0.3120 0.2205 0.2742 57 0.3797 0.2644 0.3657 0.3112 0.2184 0.2736 58 0.3810 0.2642 0.3669 0.3096 0.2149 0.2732 59 0.3823 0.2641 0.3681 0.3080 0.2131 0.2716 60 0.3836 0.2639 0.3692 0.3065 0.2114 0.2692 61 0.3849 0.2637 0.3704 0.3048 0.2095 0.2679 62 0.3861 0.2635 0.3716 0.3040 0.2079 0.2653 63 0.3874 0.2633 0.3727 0.3025 0.2058 0.2627 64 0.3886 0.2631 0.3738 0.3002 0.2043 0.2620 65 0.3899 0.2629 0.3749 0.2978 0.2024 0.2601 66 0.3911 0.2627 0.3760 0.2959 0.2010 0.2589 67 0.3923 0.2625 0.3771 0.2941 0.2000 0.2567 68 0.3936 0.2622 0.3782 0.2923 0.1989 0.2560 69 0.3948 0.2619 0.3793 0.2897 0.1971 0.2540 70 0.3961 0.2617 0.3804 0.2877 0.1955 0.2533 71 0.3973 0.2614 0.3816 0.2859 0.1928 0.2507 72 0.3986 0.2610 0.3827 0.2840 0.1914 0.2492 73 0.3999 0.2607 0.3838 0.2837 0.1896 0.2474 74 0.4012 0.2603 0.3849 0.2822 0.1878 0.2461 75 0.4025 0.2600 0.3861 0.2800 0.1860 0.2435 76 0.4038 0.2596 0.3872 0.2795 0.1844 0.2418 77 0.4051 0.2591 0.3883 0.2780 0.1834 0.2397 78 0.4065 0.2587 0.3893 0.2771 0.1816 0.2390 79 0.4079 0.2582 0.3905 0.2746 0.1789 0.2383 80 0.4093 0.2577 0.3916 0.2732 0.1773 0.2380 81 0.4107 0.2572 0.3927 0.2720 0.1756 0.2370 82 0.4122 0.2566 0.3938 0.2725 0.1730 0.2355 83 0.4136 0.2560 0.3949 0.2723 0.1707 0.2349 84 0.4151 0.2553 0.3960 0.2738 0.1687 0.2350 85 0.4166 0.2546 0.3971 0.2742 0.1651 0.2349 86 0.4181 0.2539 0.3982 0.2749 0.1627 0.2361 87 0.4198 0.2530 0.3993 0.2774 0.1590 0.2364 88 0.4215 0.2521 0.4005 0.2793 0.1556 0.2372 89 0.4233 0.2511 0.4015 0.2819 0.1523 0.2392 90 0.4252 0.2500 0.4025 0.2847 0.1492 0.2423 91 0.4273 0.2486 0.4036 0.2926 0.1463 0.2468 92 0.4294 0.2472 0.4046 0.2956 0.1443 0.2510 93 0.4318 0.2455 0.4055 0.3006 0.1419 0.2563 94 0.4344 0.2434 0.4063 0.3101 0.1427 0.2614 95 0.4373 0.2407 0.4069 0.3179 0.1432 0.2699 96 0.4403 0.2373 0.4071 0.3301 0.1451 0.2797 97 0.4438 0.2324 0.4062 0.3406 0.1474 0.2921 98 0.4475 0.2248 0.4029 0.3683 0.1521 0.3094 99 0.4475 0.2081 0.3885 0.4131 0.1712 0.3581 Finally, note that Corak and Heisz (1999) also examined the nature of nonlinearities in intergenerational earnings persistence in Canada. They estimated a flexible nonparametric model on the older version of the IID data. While a direct comparison with this study is not possible as a result of the different functional forms employed, it is interesting to note that Corak and Heisz (1999) also showed a considerable rise in the estimated elasticity at the very top of the distribution—above the 9th percentile. Unlike the results from this study, however, they found less persistence just below the 99th percentile compared with the IGE estimated for most of the upper part of the distribution. It is difficult to say whether this difference stems from the different functional forms employed in the two studies, or from the improved dataset used in this study, which better accounts for lifecycle variation in earnings. However, the results depicted in Chart 9 suggest that inaccurate proxies for lifetime earnings can significantly distort the estimated patterns of nonlinearity in the intergenerational transmission of earnings. ## 6 Conclusion Understanding the extent of intergenerational earnings and income mobility is informative for economic and social policy. In particular, estimates of intergenerational income elasticity (IGE) are often seen as broad indicators of equality of opportunity. However, because full career histories of parents and children are generally not available in the data, many existing IGE estimates for various countries may be affected by biases arising from inadequate proxies for lifetime earnings. This paper re-examines the extent to which lifecycle variation and errors-in-variables can bias IGE estimates, both at the mean of and across the percentiles of the income distribution. It is expected that the new augmented Intergenerational Income Database from Canada, with nearly full history of career data, will serve to advance understanding of this subject. The analysis shows that lifecycle bias may be present at any stage of the working career. The bias tends to be higher when annual earnings from early career years are used to proxy for lifetime earnings as the dependent variable. This finding is in line with the existing international evidence. However, the study also identified cross-country differences with respect to the age at which lifecycle bias may be minimal. This finding highlights the difficulty in making appropriate international comparisons, even when accounting for the sample and methodological differences. That is, cross-national variation in IGE estimates may still arise, as a result of differences in the age at which sons' earnings are observed. The intergenerational earnings elasticity for Canada is estimated at about 0.32. This is higher than the approximate 0.2 estimate obtained in Canadian literature. Accounting for lifecycle bias in the early estimates explains about two-thirds of the discrepancy, while errors-in-variables-induced bias contributes to the remaining difference. The study shows that the extent of the bias is larger for market and total income than for earnings alone. The results also reveal significant gender differences with regard to the effect of these biases on the estimated IGEs. The father–daughter elasticity remains quite modest irrespective of the ages at which daughters' earnings are measured. The lower IGE for daughters may be driven by gender differences in labour force participation and/or by estimation issues related to marital sorting. This highlights the need for a closer look into these patterns for future research. Using data less affected by measurement error compared with those used in previous studies, this paper explains a distinct pattern of nonlinearity in the intergenerational transmission of income in Canada. The results indicate that the relationship between fathers' and sons' earnings exhibits a rather convex pattern; one that is similar to that found in Nordic European studies, but that is in contrast to the linear pattern observed in the U.S. literature. Both polynomial and quantile regressions reveal high mobility rates at the very bottom of the distribution and low mobility at the top. One possible conjecture is that social institutions may help explain these findings. Moreover, the study demonstrates that the patterns of nonlinearity can be significantly misread when the lifecycle and errors-in-variables biases are not adequately addressed, especially over the upper part of the distribution. ## Appendix Appendix Table 1 Age distribution of fathers in 1978, Intergenerational Income Database Table summary This table displays the results of Age distribution of fathers in 1978, Intergenerational Income Database. The information is grouped by Age in 1978 (appearing as row headers), Distribution and Mean earnings over 1978 to 1982, calculated using number, percent and 2010 constant dollars units of measure (appearing as column headers). Age in 1978 Distribution Mean earnings over 1978 to 1982 number percent 2010 constant dollars 34 years of age or less 27,001 5.34 46,557 35 to 40 years of age 137,263 27.13 53,509 41 to 45 years of age 143,230 28.31 55,282 46 to 50 years of age 108,582 21.46 54,202 51 years of age and older 89,910 17.77 49,252 Total 505,986 100.00 53,155 Appendix Table 2 Illustration of sample section criteria for fathers — Linked earnings years for fathers, 1978 to 1999 Table summary This table displays the results of Illustration of sample section criteria for fathers — Linked earnings years for fathers, 1978 to 1999. The information is grouped by Name (appearing as row headers), Father's birth cohort, Father aged 35, Father aged 55, Age of father (years), Number of records from age 35 to 55, Number of records with earnings = less than$500 and Records used, calculated using 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54 and 55 units of measure (appearing as column headers).
Name Father's birth cohort Father aged 35 Father aged 55 Age of father (years) Number of records from age 35 to 55 Number of records with earnings >=\$500 Records used
35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55
A 1925 1960 1980 Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 3 3 No
B 1935 1970 1990 Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 13 13 Yes
C 1940 1975 1995 Note ...: not applicable Note ...: not applicable Note ...: not applicable raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 EApp2 - Note 2 EApp2 - Note 2 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 18 16 Yes
D 1943 1978 1998 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 21 21 Yes
E 1944 1979 1999 raApp2 - Note 1 raApp2 - Note 1 EApp2 - Note 2 raApp2 - Note 1 raApp2 - Note 1 EApp2 - Note 2 EApp2 - Note 2 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 EApp2 - Note 2 raApp2 - Note 1 raApp2 - Note 1 EApp2 - Note 2 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 EApp2 - Note 2 21 15 Yes
F 1950 1985 2005 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable 15 15 Yes
G 1950 1985 2005 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 EApp2 - Note 2 EApp2 - Note 2 EApp2 - Note 2 EApp2 - Note 2 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 EApp2 - Note 2 EApp2 - Note 2 raApp2 - Note 1 EApp2 - Note 2 EApp2 - Note 2 Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable 15 7 No
H 1955 1990 2010 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable 10 10 Yes
I 1960 1995 2015 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 raApp2 - Note 1 Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable Note ...: not applicable 5 5 No

## References

Aina, C., and C. Nicoletti. 2013. The intergenerational transmission of liberal professions: nepotism versus abilities. Unpublished mimeo.

Baker, M., and G. Solon. 2003. "Earnings Dynamics and Inequality among Canadian Men, 1976-1992: Evidence from Longitudinal Income Tax Records." Journal of Labor Economics 21 (2): 289–321.

Becker, G.S., and N. Tomes. 1986. "Human Capital and the Rise and Fall of Families." Journal of Labor Economics 4 (3): S1–S39.

Björklund, A. 1993. "A Comparison between Actual Distributions of Annual and Lifetime Income: Sweden 1951-89." Review of Income and Wealth 39 (4): 377–386.

Björklund, A., and M. Jäntti. 2010. "Intergenerational Income Mobility and the Role of Family Background." In Oxford Handbook of Economic Inequality, ed. W. Salverda, B. Nolan, and T. Smeeding. Oxford: Oxford University Press.

Björklund, A., J. Roine, and D. Waldenström. 2012. "Intergenerational top income mobility in Sweden: Capitalistic dynasties in the land of equal opportunity?" Journal of Public Economics 96 (5-6): 474–484.

Böhlmark, A., and M.J. Lindquist. 2006. "Life-Cycle Variations in the Association between Current and Lifetime Income: Replication and Extension for Sweden." Journal of Labor Economics 24 (4): 879–900.

Bratsberg, B., K. Røed, O. Raaum, R. Naylor, M. Jäntti, T. Eriksson, and E. Österbacka. 2007. "Nonlinearities in Intergenerational Earnings Mobility: Consequences for Cross-Country Comparisons." The Economic Journal 117: C72–C92.

Brenner, J. 2010. "Life-Cycle Variations in the Association between Current and Lifetime Earnings: Evidence for German Natives and Guest Workers." Labour Economics 17 (2): 392–406.

Chadwick, L., and G. Solon. 2002. "Intergenerational Income Mobility among Daughters." The American Economic Review 92 (1): 335–344.

Chetty, R., N. Hendren, P. Kline, and E. Saez. 2014. "Where is the Land of Opportunity? The Geography of Intergenerational Mobility in the United States." The Quarterly Journal of Economics 129 (4): 1553–1623.

Corak, M. 2001. "Are the Kids All Right? Intergenerational Mobility and Child Well-Being in Canada." In Review of Economic Performance and Social Progress, ed. K. Banting, A. Sharpe and F. St-Hilaire, p. 273–292. Montréal and Ottawa: Institute for Research on Public Policy and Centre for the Study of Living Standards.

Corak, M. 2006. "Do Poor Children Become Poor Adults? Lessons from a Cross Country Comparison of Generational Earnings Mobility." In Research on Economic Inequality, ed. J. Creedy and G. Kalb, vol. 13 (1), p. 143–188. Amsterdam: Elsevier B.V.

Corak, M. 2013. Income Inequality, Equality of Opportunity, and Intergenerational Mobility. IZA Discussion Paper no. 7520. Bonn: The Institute for the Study of Labor (IZA).

Corak, M., B. Gustafsson, and T. Österberg. 2004. "Intergenerational Influences on the Receipt of Unemployment Insurance in Canada and Sweden." In Generational income mobility in North America and Europe, ed. M. Corak. Cambridge, U.K.: Cambridge University Press.

Corak, M., and A. Heisz. 1999. "The Intergenerational Earnings and Income Mobility of Canadian Men: Evidence from Longitudinal Income Tax Data." The Journal of Human Resources 34 (3): 504–533.

Corak, M., M. Lindquist, and B. Mazumder. 2014. "A Comparison of Upward and Downward Intergenerational Mobility in Canada, Sweden and the United States." Labour Economics 30 (C): 185–200.

Corak, M., and P. Piraino. 2011. "The Intergenerational Transmission of Employers." Journal of Labor Economics 29 (1): 37–68.

Duncan, G.J., and R.J. Murnane. 2011. "Introduction: The American Dream, Then and Now." In Whither Opportunity? Rising Inequality, Schools, and Children's Life Chances, ed. G.J. Duncan and R.J. Murnane. New York: Russell Sage Foundation.

Dunn, T., and D. Holtz-Eakin. 2000. "Financial Capital, Human Capital, and the Transition to Self-Employment: Evidence from Intergenerational Links." Journal of Labor Economics 18 (2): 282–305.

Ermisch, J., M. Francesconi, and T. Siedler. 2005. Intergenerational Economic Mobility and Assortative Mating. Discussion Paper no. 1847. Bonn: The Institute for the Study of Labor (IZA).

Fortin, N., and S. Lefebvre. 1998. "Intergenerational Income Mobility in Canada." In Labour Markets, Social Institutions, and the Future of Canada's Children, ed. M. Corak. Statistics Canada Catalogue no. 89-553-X. Ottawa: Statistics Canada.

Gouskova, E., N. Chiteji, and F. Stafford. 2010. "Estimating the intergenerational persistence of lifetime earnings with life course matching: Evidence from the PSID." Labour Economics 17 (3): 592–597.

Grawe, N.D. 2004. "Reconsidering the Use of Nonlinearities in Intergenerational Earnings Mobility as a Test for Credit Constraints." Journal of Human Resources 39 (3): 813–827.

Grawe, N.D. 2006. "Lifecycle bias in estimates of intergenerational earnings persistence." Labour Economics 13 (5): 551–570.

Haider, S., and G. Solon. 2006. "Life-Cycle Variation in the Association between Current and Lifetime Earnings." The American Economic Review 96 (4): 1308–1320.

Han, S., and C.B. Mulligan. 2001. "Human Capital, Heterogeneity and Estimated Degree of Intergenerational Mobility." The Economic Journal 111 (470): 207–243.

Kramarz, F., and O.N. Skans. 2014. "When Strong Ties are Strong: Networks and Youth Labour Market Entry." The Review of Economic Studies 81: 1164–1200.

Lentz, B.F., and D.N. Laband. 1990. "Entrepreneurial Success and Occupational Inheritance among Proprietors." Canadian Journal of Economics 23 (3): 563–579.

Mazumder, B. 2005. "Fortunate Sons: New Estimates of Intergenerational Mobility in the United States Using Social Security Earnings Data." The Review of Economics and Statistics 87 (2): 235–255.

Mulligan, C.B. 1997. Parental Priorities and Economic Inequality. Chicago: University of Chicago Press.

Nybom, M., and J. Stuhler. 2011. "Heterogeneous Income Profiles and Life-Cycle Bias in Intergenerational Mobility Estimation." Discussion Paper no. 5697. Bonn: The Institute for the Study of Labor (IZA).

Oreopoulos, P. 2003. "The long-run consequences of growing up in a poor neighborhood." Quarterly Journal of Economics 118: 1533–1575.

Oreopoulos, P., M. Page, and A.H. Stevens. 2008. "The intergenerational effects of worker displacement." Journal of Labor Economics 26: 455–484.

Page, M.E. 2004. "New Evidence on Intergenerational Correlations in Welfare Participation." In Generational income mobility in North America and Europe, ed. M. Corak. Cambridge, U.K.: Cambridge University Press.

Pérez-González, F. 2006. "Inherited Control and Firm Performance." American Economic Review 96 (5): 1559–1588.

Raaum, O., B. Bratsberg, K. Røed, E. Österbacka, T. Eriksson, M. Jäntti, and R.A. Naylor. 2007. "Marital sorting, household labor supply and intergenerational earnings mobility across countries." The B.E. Journal of Economic Analysis and Policy 7 (2): Article 7.

Roemer, J.E. 1998. Equality of opportunity. Cambridge, Massachusetts: Harvard University Press.

Statistics Canada. 2010. Intergenerational Income Data Reference Manual. Unpublished manuscript. Ottawa.

Solon, G. 1992. "Intergenerational Income Mobility in the United States." American Economic Review 82 (3): 393–408.

Solon, G. 1999. "Intergenerational Mobility in the Labor Market." In Handbook of labor economics, ed. O.C. Ashenfelter and D. Card, vol. 3A, p. 1761–1800. Amsterdam: Elsevier Science B.V.

Solon, G. 2002. "Cross-country differences in intergenerational earnings mobility." Journal of Economic Perspectives 16 (3): 59–66.

Sorensen, J.B. 2004. Closure and Exposure: Mechanisms in the Intergenerational Transmission of Self-employment. Unpublished mimeo.

Is something not working? Is there information outdated? Can't find what you're looking for?