Income Research Paper Series
Low Income in Canada - A Multi-line and Multi-index Perspective
Low-income trends in Canada
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.
This chapter attempts to answer the question: what are the major trends in low income in Canada over the last few decades? At Statistics Canada, a historically consistent income data series allows us to examine low income over a period of 34 years, from 1976 to 2009. The period witnessed some important structural changes in the Canadian economy due to, most notably, the introduction of new information and communication technologies, the establishment and enlargement of the free trade zones, globalization and increasing international competition. The decades covered two complete economic cycles in which both GDP and employment contracted in absolute terms. A new contraction started in late 2008. It was also a period in which redistribution policies such as employment insurance and social assistance programs evolved dramatically. All of these changes have implications for low income and this is the first study that documents low income for such a long period using multiple measures and indicators.
The evolution of low income incidence: 1976 to 2009
The low-income rate, also referred to as the incidence of low-income, tells us what proportion of a population is living with an income that is below a given threshold.1 Figure 2.1 presents the low income rate (left axis) for all Canadians from 1976 to 2009. It also shows how the unemployment rate (right axis, for Canadians aged 15 to 64 years old) evolved during that period. The figure suggests that while the low-income rates in Canada varied significantly over time, they roughly followed the ups and downs of the overall economy and sometimes low income measured by different lines tell somewhat different stories.
Figure 2.1 clearly shows that the low- income rates measured by fixed historical standards such as the Low Income cut-offs (LICO) and the Market Basket Measure (MBM) varied more over time than did the Low Income Measure (LIM) which is based on standards of current income distributions. Under LICO, the low-income rate reached a historical high of almost 16% in 1996, and about a decade later, a historical low of just above 9% was observed. There was also another pair of high and low points in low income rate under LICO: the 1983 high of 14% and the subsequent low of 10% in 1989. Both comparisons are statistically significant.2 The MBM line has a relatively short history, but the historical low under MBM coincided with that under LICO. Under LIM, however, the historical low incidence coincided with the 1989 low of the LICO and the low income rate under LIM in that year was significantly lower than it had been or would be for any other year during the last three decades.
Figure 2.1 Low-income rates and unemployment rate, Canada, 1976 to 2009
The short-term variations in the low-income rates appeared to follow business cycles closely. Before the year 2000, low income incidences measured by the LICO and the LIM seemed to evolve along the same path. This path coincided with changes in the unemployment rate. For example, the early 1980s recession was immediately followed by a high of low income incidence under both LICO and LIM, varying between 13 to 14% in the period of 1983-1984. It then dropped quickly to a low of 10% by 1989 under both lines.
However the variations in low income did not follow the variations in unemployment all the time. For example, from 1976 to 1981, low income rates declined while the unemployment rate fluctuated around 8%. Likewise, from 1993 to 1996, while unemployment rate dropped by almost two percentage points, low income rates under both LICO and LIM increased, and the increase in low income rate under LICO was more than one percentage point. These observations suggest that labour market is an important factor behind the evolution of low income incidence, but not the only one. The evolution of income, particularly the evolution of income at the lower end of the distribution is another important factor.
Figure 2.2 shows how individual after-tax income (in current dollars) evolved between 1976 and 2009. The figure indicates that change in income is an important factor behind the low income trend. We can divide the whole period into three sub-periods. From 1976 to 1988, the median after-tax income increased strongly. The annualized rate of increase amounted to 8.3%. But from 1989 to 1995, the median after-tax income increased only by an annualized rate of 1.1%. While from 1996 to 2009, the median increased by an annualized rate of 4.0%. This evolution of individual after-tax income coincided with the decline in the low income rates from 1976 to the late 1980s, the increase from the late 1980s to the mid 1990s, and the decline (at least under LICO and MBM) during the last decade.
Figure 2.2 Median after-tax income 1976 to 2009
Figure 2.3 provides further details for the evolution of individual after-tax income by income quartiles. The income growth rates for those who were in the lower portion of the distribution were closely tied to the evolution of the low income rates. During the first sub-period (1976-1988), median after-tax income for those who were in the bottom quartile increased much faster than for those who were in the top quartiles. The annualized growth rates for the two groups were 16.8% and 6.6%, respectively. While for those who were in the second and third quartiles, the annualized growth rates were 10.1% and 7.3%. From the 1989 to 1995, median income for those who were in the bottom quartile actually declined slightly while that for the top quartile continued to increase at an annual rate of 6.1% and hence, the low income rates in this period increased. Finally, from 1996 to 2009, income in the bottom quartile again increased faster than income in the top, at 4.4% and 3.7%, respectively. However, in this period, low income rates under LICO (and later under MBM) declined substantially, while rates under LIM did not, suggesting that there might be other factors at play.
Figure 2.3 Income growth paths by income quartile
One factor is the way low income is determined. The real values of the low income thresholds under LICO and MBM are fixed in time. The changes of those thresholds over time reflect purely the change in inflation. For example, the current LICO thresholds were set according to the 1992 consumption pattern. For the same person living in the same place in 1992 and 2009, the LICO thresholds were the same in terms of real values. They differed only by a nominal value. On the other hand, the LIM thresholds are updated each year according to the contemporary income distribution, and hence, when income increases, the threshold is automatically increased. However under the LIM current income is compared with a current standard, and as such the low income status of a person is independent of inflation.3
Figure 2.4 illustrates how average after-tax income and the LICO and LIM thresholds grew over time.4 From 1976 through the 1990 all thresholds grew at similar rates and the LICO and LIM incidences followed a similar pattern (see Figure 2.1). The figure also shows how the Consumer Price Index (CPI) evolved over time. Through the 1990s, the CPI grew faster than average incomes and consequently the LICO thresholds increased more quickly than the LIM thresholds, resulting in a sharper increase in low income rates under LICO than under LIM. However, starting from the late 1990s, median incomes began to grow faster than did the CPI and so the LIM thresholds increased faster than did the LICO thresholds. The result was a significant drop in the low income rate under LICO while the low income rate under LIM showed stability with a slightly upward trend.
Figure 2.4 Growth paths of average income and LICO/LIM thresholds (2001=1)
Of course, the overall picture of low income incidence does not tell the whole story. On the one hand, there are other low income indexes that may provide us with additional information. This will be examined later in this chapter. On the other hand, the composition of the low income population may also change over time, and among those who were in low income, some might have done better than others. The evolution of low income for different groups of individuals, including those from different regions will be examined in later chapters.
Nevertheless, the evolution of the low income rates in 1976 to 2009 suggests that different measurement schemes may, at times, tell a different story. In particular, from 1996 to 2009, while the low income rates under LICO (and later under MBM) declined substantially, the low income rate under LIM did not change much. This raises the issue of how often one should rebase the low income threshold. If we use a fixed standard such as LICO, we are comparing the income of those who are in the lower part of the income distribution today with those who were in the lower portion of the distribution in 1992. In this way, we may measure progress against earlier situations. But if we use a variable standard such as LIM, each year we compare the income of those who are in the lower portion of the distribution today with those who are in today's middle. In other words the LICO and MBM tend to track change relative to past standards while the LIM tracks the low income population relative to the current distribution.
The report now turns to examine how low income changed following the latest recession.
How did low-income rates change during the recent recession?
Before the global financial crisis hit in 2008, Canada had enjoyed more than ten years of economic growth and rising employment. But the sudden reduction in world demand in the summer of 2008 quickly spread into the goods production and the labour market in Canada. Goods production posted the first decline since the 2000 recession; employment fell by 216,000 in agriculture, construction and manufacturing sectors in the last quarter of 2008. The situation deteriorated further in 2009. For example, the unemployment rate among Canadians aged 15 to 64 reached 8.4%--the highest level in the last decade. Given these, it is important to see what happened to low income during the recession.
Under the low income cut-offs (LICO), 9.4% and 9.6% of Canadians were in low income in 2008 and 2009, respectively (Table 2.1). Compared to the years from 2000 to 2006, the low-income rate in 2009 was significantly lower; but compared to 2007, the low-income rate under LICO increased by two-tenths and four-tenths of a percentage point in 2008 and 2009 but these changes are not statistically significant, as shown by Figure 2.5 (Panel A).5
However, as discussed before, LICO is only one of the available low income thresholds in Canada. What does the picture look like if we take a snap shot from a different angle? Let's examine low income using the Market Basket Measure first. Table 2.1 indicates that under MBM, the low-income rates in 2008 and 2009 were 9.5% and 10.6%. These are still significantly lower than that in 2000, but compared to 2007, the low-income rate in 2009 was significantly higher, while the low income rate in 2008 was not statistically different from that in 2007 (Figure 2.5, Panel C).
Table 2.1. Low income rates under different lines
Under LIM, as can be seen from Table 2.1, the low-income rates were 13.3% in 2008 and 2009. This means that 13.3% of Canadians had an income below half of the median of the adjusted national income distribution.6 The LIM threshold tells us how the income of an individual from the lower portion of the distribution compares with that of the individual from the middle of the distribution. It shows that the low-income rates increased by one eighth of a percentage point between 2007 and 2008/2009. But in this case, the change was not significant. Indeed, under LIM, annul low income rate remained virtually the same during the 2000-2009 period (Figure 2.5, Panel B).
Figure 2.5 The 95% Confidence interval estimates of low-income rates under different lines
Apparently, different lines provide us with different answers to the question as whether low income increased, decreased or was unchanged in 2008/2009. If we were to rely on the traditional LICO line only, we would have inferred that the observed four-tenths of a percentage point increase in the low-income rates between 2007 and 2009 can be attributed to sampling errors and is negligible. But under MBM, we saw evidence that the low-income rates had increased in 2008/2009 and an upward trend in low income was emerging.
How robust is this inference? After all, the confidence interval estimates on low-income rates under LICO and LIM for the two years still overlapped, and since each low-income line is subject to its own assumptions and choices, it might well be the case that the estimated changes in the low-income rates under different lines were biased. To deal with this ambiguity, we need a more general tool. Instead of employing one or several thresholds, we consider a range of possible choices of low income thresholds and ask the question: how would the low-income rates change if the threshold is set at $1,000, $2,000, $3,000, ..., $30,000 per person per year? If, for a variety of reasonable thresholds, we also observe that the low-income rate in 2009 was higher than in 2007, we would be able to make a more robust conclusion.
We calculated low-income rates for 2007, 2008 and 2009 for low-income thresholds varied from $100 to $30,000 per person/year.7 Figure 2.6 illustrates the estimated low-income rates for low-income lines ranged between $10,000 and $18,000. Our calculation shows that, for a range of low-income line from $11,000 to $17,000, the low-income rate in 2008 was higher than in 2007. That is, if it can be agreed that the low-income line should be set somewhere between $11,000 and $17,000, then we would be able to conclude that the low income rate increased between 2007 and 2008.8 The figure also indicates that low income rate in 2009 was higher than that in 2007 if the low income line is set between 12,500 and 14,500 and between 15,000 and 17,000. There is, however, some ambiguity when the line is set between 14,500 and 15,000.
We conducted further tests using a series of variable low-income lines under the LIM methodology. We found that when the low income lines were set between 35% to 75% of the medians of the 2007 and 2009 income distributions, the low-income rate in 2009 was higher than in 2007, and the differences were statistically significant. Other tests show similar results.9 Therefore, there is evidence to suggest that low income in 2008/2009 had a small increase relative to that in 2007.
Figure 2.6 Low income rates in 2007, 2008 and 2009 under a range of low income thresholds
Our observations thus far are based on multiple low income lines. To be consistent with best practices, we also need to examine the evolution of low income with multiple indexes. The results are presented in the next section.
Low-income gap ratio and severity
The low-income rate is a very straightforward index and while simple to understand it has certain limits. For example, the inequality among low-income persons is also a concern for policy makers yet the low-income rate does not contain any information about that inequality. Indeed, "(S)ince the work of Sen (1976), taking into account inequality among the poor, and not solely the incidence or average intensity of poverty, has become common scientific practice".10 The statement equally applies to studies of low income. To be consistent with best practices, this section examines how low-income gap ratio and severity have evolved in Canada.11
The evolution of the low-income gap ratio in Canada during the 1976 to 2009 period is presented in Figure 2.7. The figure appears to convey the same trend as that by Figure 2.1: the low income gap ratio changed in the same direction as the low-income rate in the whole period and the variations in the gap ratio under LICO were much stronger than under LIM.
Figure 2.7 Low-income gap ratios under alternative lines
But some new insights can also be seen by examining the gap ratio statistics. For example, even though the low-income rates changed little in the years between 2001 and 200612, no matter which line was employed (Figure 2.5), when the gap ratio statistics are examined, the estimates suggest that the gap ratio in 2006 was significantly lower than in 2004 under LIM. The gap ratio was also significantly lower in 2006 than in 2001 under MBM, and more remarkably, the gap ratio in 2006 was lower than those in all the other five years under LICO. Consequently the stable low-income rate during this period appeared to be accompanied by reductions in the depth of low-income, and this is reinforced by the observation that the low income gap ratio declined under all three lines between 2000 and 2007.
Our results also show that the estimated low income gap ratios for 2009 (and to a certain extent for 2008) under both LIM and MBM were significantly higher than those in 2007. This corroborates our earlier finding that the low income rate in 2009 increased relative to 2007.
The low-income severity index measures how unequal low-income persons are between themselves. One measure of this is the square of the gap ratio. Under this index, low-income persons with large income shortfalls contribute more than low-income persons who have smaller shortfalls. In addition to the monotonic axioms, the index satisfies the transfer axiom: other things being equal, a transfer of income from one low-income person to anybody with higher income must increase the severity of low-income.13
Figure 2.8 Low-income severity indexes under different lines, Canada
To understand the measure of severity, consider an extreme case in which we take away an amount from a person with the lowest income and transfer the money to a person who is just below the low income line, but the amount transferred is not large enough to lift the recipient out of low income. Such a transfer would not change the low-income rate or the gap ratio, but the low-income severity index would increase and hence it is the severity index that will be able to capture the increase in inequality among low-income persons.
The evolution of low-income severity indexes from 1976 to 2009 is presented in Figure 2.8. The general trend in low-income severity indexes under different lines is similar. They generally declined from 1976 to 1989, then increased until the mid to late 1990s and decreased again thereafter. The latest declining trend started to reverse in 2008, confirming the emerging upward trends in both the low-income rate and the gap ratio statistics discussed earlier. As in the case for the low-income gap ratio, we also found that the severity indexes in 2009 were significantly higher than in 2007 under both the LIM and the MBM, but not under the LICO.
Interestingly, we found that, under LIM, low income severity indexes in 2006 and 2007 were all significantly lower than in 2000, 2004 and 2005. This is in sharp contrast to the finding that low-income rates under LIM remained unchanged in the 2000 to 2009 period. These results demonstrate the importance of examining different low-income indexes in order to obtain a fuller picture of individuals' well-being.
From 1976 through 2009 the low-income rate, gap ratio and severity indexes under the three low-income lines varied significantly, but they generally evolved along the same cyclical trend. However, trends in low-income rates under different lines sometimes diverged. In particular, for the 2000-to-2009 period, we found the low-income rate declined under LICO and MBM, the fixed standards, while under LIM, the relative standard, no statistically significant change was observed. This would suggest that progress was made against a historical standard but that relative to current populations, low income incidence was stable. On the other hand, the evolution of the low income gap ratio and the severity indexes suggests that the well-being of Canadians from the bottom of the distribution had improved under both fixed and relative standards before the start of the recent recession.
Did low-income increase in 2008/2009 relative to 2007? Our results show that if a single line such as the LICO is employed, the answer would be no. But by examining multiple lines and multiple indexes, a more complete picture emerged. Our analysis indicates that an uptick in low income occurred in 2008/2009, and results based on more general statistical tests confirmed that the low-income rate increased slightly between 2007 and 2008/2009 under a range of possible low income thresholds.
1. We use the after-tax income concept under LICO and LIM and the disposable income concept under MBM in this report. After-tax income is total household/economic family income minus income taxes. The disposable income is defined as income available to purchase the goods and services that are contained in the MBM basket. In particular, MBM's disposable income = total income – income taxes – CPP/QPP contribution – EI contribution – RPP contribution – union dues – child/spousal support payment – work-related child care expenses – out of pocket medical expenses – public health insurance premiums.
2. Statistical tests for the period before 2000 are based on estimated standard errors with analytical formulae for the FGT indexes. For the period thereafter in which all three low income lines are available and on which our analysis was focused on, bootstrap weights from SLID were employed. A comparison of the results under the two approaches suggests that estimates using the bootstrap weights are more conservative. For example, low income rate under LIM in 2008 was significantly higher than that in 2007 under the first approach. But under the second approach, the difference was only marginal.
3. The assumption is that everyone faces the same prices
4. The mean threshold and income figures are calculated by averaging across individuals the value of the family threshold or income. The growth rates have been normalized to 2001. We performed robustness checks to confirm that the year of normalization did not bias the results shown here.
5. We determine statistic significance by the 95% confidence interval estimate of the low-income rate. When the interval estimates for two different years do not overlap, we say that low-income rates in these two years are significantly different, or the change in low-income rates between the years is statistically significant. If the intervals overlap completely, we say that change is not significant, and when they overlap somewhat, we would say that the change is only marginally significant. Notice that the SLID samples from the same panel and some of those from two adjacent panels are not independent and thus the estimated low income indexes are correlated. In this report, these situations are taken into consideration to ensure the validity of our inferences.
6. Please see the Methodology Appendix for LIM for detail.
7. Income is measured in 2009 constant dollar. Household is used as the unit to determine the low-income status of an individual and the square root of household size is used as the equivalence scale. We also tested with economic family as the unit for low-income determination. The result is the same.
8. To give our reader some ideas for different lines, we notice that in 2008, the LIM threshold for a single person was $18,582; the LICO threshold for a single person living in an urban area with 500,000 or more residents was $18,373; while the MBM threshold for a person lived in Toronto was $15,565. Using CPI to adjust Professor Sarlo's threshold ($8,900 in 2000), one would obtain $10,644 for 2008.
9. Through stochastic dominance tests (with DAD 4.5 by Duclos, Araar and Fortin 2008), we found the 2007 distribution dominates that of 2009 in first order for thresholds blow $25,575. The 2007 distribution also dominates that of 2008 in first order for thresholds below $19,995 (all in 2009 constant dollars).
10. Duclos and Grégoire (2003).
11. Additional discussions on the low-income gap ratio and severity indexes can be found in the Methodology Appendix.
12. The confidence interval estimates largely overlapped each other in these years.
13. The axiomatic approach is pioneered by Sen (1976) and examined by many authors in the poverty literature.