# Section 6: The many factors leading to low earnings: a multivariate analysis

The logistic regression analysis reported in this section first considers the contribution of demographic or 'given' characteristics on the probability that an individual will have employment earnings at or below the national median employment earnings in 2006. Other groups of variables are then added progressively in order to assess both their independent effects and whether they modify the effects of previously-added variables. Thus, after the demographic variables, education variables are added and finally, the work-related variables. The logic behind this approach is that one is born with certain characteristics, then one goes to school (and this has an effect on earnings) and then finally one enters the labour market.

The analysis is first performed on the entire population aged 25 to 64 who had non-zero employment earnings in 20061 in order to measure the relative effect of different levels of education on one's probability of falling into low earnings. The second part of the analysis focuses on college and university graduates in order to assess the effect of field of study.

## Box 6.1 Logistic regression analysis

Logistic regression analysis produces odds ratios, which, in this study, are used to assess whether, other things being equal, workers with specific characteristics are more or less likely to fall into low earnings than those in another (reference) group. For example, consider the risk of low earnings for a woman as compared to a man. An odds ratio near 1.0 implies that the two groups have the same odds of falling into low earnings; an odds ratio that is less than 1.0 implies that those in the group being considered are less likely to fall into the low earnings category than the reference group; an odds ratio greater than 1.0 implies that those in the group being considered are more likely to fall into low earnings than those in the reference category. To illustrate, consider two scenarios: 1) females being the reference category and males having an odds ratio of 0.65, and 2) females being the reference category and males having an odds ratio of 1.75. The first scenario indicates that males are 0.65 as likely as females to have low earnings, whereas the second scenario indicates that males are 1.75 times as likely as females to have low earnings.

In the analysis reported here, the dependent variable equals 1 if an individual has employment earnings at or below half the national median employment earnings in 2006 and 0 otherwise.

## 6.1 The total population – multivariate analysis

### Demographic factors

The first-order relationship between gender and the likelihood of falling into low earnings was strong. When only controlling for demographic characteristics such as age, family situation and immigration status, men were slightly less than half as likely as women to fall into a situation of low earnings (Table A.11, Column 1). This relationship was barely affected by the addition of education variables (Table A.11, Column 2) and was slightly weakened by the addition of information about major activity (Table A.11, Column 3). That the gap between men and women decreased when major activity was controlled for is not surprising, given that previous literature has found that women's earnings are affected by caregiving responsibilities, either for children or for aging parents. This was also true once occupation and industry were taken into account (Table A.11, Column 4), though the addition of occupation and industry did have the effect of slightly decreasing the difference between men and women in the likelihood of falling into low earnings. Previous research has found that women tend to work in lower-paying industries and occupations more than men; controlling for these differences results in a decrease in the gender gap in earnings.

Overall, gender retained a strong relationship with the likelihood of falling into low earnings, even after controlling for other factors. Thus, in the final model when all variables were included, gender retained an effect that was independent of other demographic factors, province, education, major activity for the year, industry, occupation and being self-employed, with women being more likely than men to fall into the lowest earnings category. That being said, the gender difference in Ontario (odds ratio for men compared to women, at 0.57; Table A.11, Column 6) was smaller than at the Canada level (odds ratio for men compared to women, at 0.49; Table B.11, Column 7).2

As was also observed in the descriptive analysis, the effects of age on the probability of falling into the lowest earnings category were polarized. In Ontario, individuals aged 25 to 29 were more likely than the control group (workers aged 44 to 49) to fall into the lowest earnings category in 2006. At the other end of the age spectrum, 55 to 59 year-olds were 1.79 times, and 60 to 64 year-olds were 3.43 times, as likely to be in the lowest earnings category compared to the reference group, before controlling for the influence of other variables (Table A.11, Column 1).

Education slightly accentuated these effects with both younger and older workers becoming more likely to fall into low earnings once education was taken into account (Table A.11, Column 2). Major activity helped eliminate the age effects for younger workers and workers aged 55 to 59 (Table A.11, Column 3). Once major activity was controlled for, neither 25 to 29 year-olds nor 55 to 59 year-olds were more likely to fall into low earnings than the control group. It is probable that the low earnings experienced by younger workers may be explained by the fact that for some, working was not their main activity for the year (that is, some were students) and some older workers were retired. It should be noted, that while the probability of falling into low earnings for workers aged 60 to 64 was reduced, it still remained higher than the control group (Table A.11, Column 3). However, the risk of falling into low earnings was pronounced for older and younger workers once controlling for the effects of industry and occupation (Table A.11, Column 4) and was even more so once self-employment status was taken into account (Table A.11, Column 5).

Once controlling for the influence of all other factors, an interesting contrast between the situation in Ontario and at the Canada level emerges. Whereas at the Canada level, individuals aged 25 to 29, 30 to 34, 55 to 59 and 60 to 64 were all significantly more likely than those in age groups ranging from 35 to 54 to be in the lowest earnings category (Table B.11, Column 7), this was not the case for Ontario, where no statistically significant differences across age groups were found (Table A.11, Column 6).

Contrasts between Ontario and the Canada level are also apparent when immigrant status is considered. In the first-order model at the Canada level (Table B.11, Column 1) immigrants were more likely than non-immigrants to fall into low earnings. This was not the case for immigrants in Ontario in the first-order model (Table A.11, Column 1). In Ontario, immigrants who had been in Canada less than 30 years only became more likely to fall into low earnings once major activity was taken into account (Table A.11, Column 3). There were no such effects when controlling for education (Table A.11, Column 2) and industry /occupation (Table A.11, Column 4). When all factors were considered together in the complete model for Ontario, immigrants who had been in Canada less than 10 years were more likely than non-immigrants to be in the lowest earnings category (odds ratio of 1.84; Table A.11, Column 6). At the Canada level, both recent immigrants (in Canada for less than 10 years) (odds ratio = 1.78) and more established immigrants (odds ratio = 1.47) were more likely than non-immigrants to be in the lowest earnings category (Table B.11, Column 7).

The Canada-level analysis finds strong effects associated with living in certain provinces. Across all model specifications, individuals in the Atlantic provinces and Quebec were much more likely to fall into low earnings than individuals in Ontario (the reference group). In contrast, individuals in Alberta were significantly less likely (odds ratio = 0.77) to be in the lowest earnings category.

### Education factors

Contrasts are also apparent between Ontario and the Canada level when level of education is considered. At the Canada level, education effects were strong in all models, although they were slightly mitigated once industry and occupation were accounted for (Table B.11, Column 5). In this case, after taking all other factors into account, those with less than high school completion were significantly more likely than high school or trades graduates to be in a low earnings situation (odds ratio = 1.44; Table B.11, Column 7) and college and university graduates were significantly less likely (odds ratios = 0.86 and 0.57, respectively).

In Ontario, however, education effects were different than those observed at the national level. In the first-order model (Table A.11, Column 1), while the familiar effects for workers who had not graduated from high school (at higher risk of falling into low earnings) and university graduates (at lower risk of falling into low earnings) were observed, there was no association with having a college diploma. These results were repeated once major activity and self-employment status were taken into account (Table A.11, Columns 3 and 5). Furthermore, the effect of having less than a high school education was no longer significant once industry and occupation were taken into account (Table A.11, Column 4). Once controlling for the effect of all factors, only university degree-holders were significantly less likely than high school or trades graduates to have low earnings (odds ratio=0.51) (Table A.11, Column 6).

### Labour market factors

Occupation is another important factor in understanding whether or not workers would fall into low earnings. In Ontario, as at the Canada level, compared to workers in sales (the reference group) workers in management, business administration, science, health, social science and the trades were all less likely to fall into low earnings (Tables A.11 and B.11). However, at the Canada level, workers who held an occupation unique to a primary industry were more likely than those in sales to fall into low earnings (and this relationship held even after controlling for other factors).

In terms of main activity and work schedule, it is not surprising to find that workers who did not work full time-full year or whose main activity during the year was something other than working were much more likely to fall into low earnings. This was the case for workers who reported that being self-employed was their main activity for the year as well. These effects were noted in Ontario and at the national level.

Finally, the industry in which a person worked had an effect on their odds of falling into low earnings. Workers in manufacturing, the finance industry, the professional, scientific and technical services industry and public administration were all less likely than workers in the health industry (the reference category) to fall into low earnings. Conversely, workers in the food industry were more likely to fall into low earnings.

It should be noted that when controlling for main activity, there was a large decrease in the odds ratios for 60 to 64 year-olds, falling from 3.43 (Table A.11, Column 1) to 1.83 (Table A.11, Column 3), indicating that for some, having 'retirement' listed as their major activity for the year explained some of the increased probability of falling into low earnings. Results were similar at the Canada level.

## 6.2 The postsecondary-educated population – multivariate analysis

In order to study the effects of field of study on the probability of falling into low earnings, it was necessary to restrict the analysis to the postsecondary-educated (that is, college and university graduates) population. Tables A.12 and B.12 show the results of this investigation.

### Demographic factors

As in the case of the analysis of the entire population, a gender effect is also observed for the postsecondary-educated population. In all models except the one controlling for main activity for the year, the odds of males being in low income were about half those of females, rising to 0.57 when all variables were taken into account. When considering the effect of main activity for the year, the analysis finds that the gender effect weakened, with the odds for males standing at 0.65 in Ontario (Table A.12, Column 4) and at 0.70 at the Canada level (Table B.12, Column 5). However, even with this weakened effect, females in Ontario and at the Canada level were still more likely than males to have low earnings.

With respect to age, in the demographic model (Table A.12, Column 1), postsecondary-educated adults aged 25 to 29 and over age 55 were more likely to fall into the lowest earnings category. This relationship remained even when controlling for whether the person had gone to college or university, their field of study, their industry and occupation and was particularly strong when controlling for self-employment status (Table A.12, Columns 2, 3, 5, and 6). As was the case for the entire population, the increased likelihood that young adults would be in the lowest income category disappeared when major activity was taken into account (Table A.12, Column 4), likely because some of them would be being students as opposed to working. Notably, the age effects disappeared for the older workers in the final model (Table A.12, Column 7) which controlled for all factors together.

Thus, in the final model, the postsecondary-educated population aged 25 to 29 was more likely than workers aged 45 to 49 (the reference category) to be in the lowest earnings category, once controlling for the impact of all other factors, with the odds being higher in Ontario (odds = 2.35, Table A.12, Column 7) than at the Canada level (odds = 1.55; Table B.12, Column 8). This was also the case for 55 to 59 and 60 to 64 year-olds at the Canada level, but not in Ontario. In Ontario, once all factors were taken into account, workers aged 55 or more were no more likely to be in the lowest earnings category than workers aged 45 to 49.

The effect of immigrant status was weaker for the postsecondary-educated population than for the general population at both the Ontario and Canada levels. At first glance, in the demographic model (Table A.12, Column 1) immigrants were no more likely than non-immigrants to fall into low earnings in Ontario. However, recent immigrants (those who had been in Canada less than 10 years) were more likely than non-immigrants to be in low earnings once educational factors such as level of education and major field of study were accounted for (Table A.12, Columns 2 and 3). This effect disappeared, however, once major activity and industry and occupation were controlled for (Table A.12, Columns 4 and 5). Notably, although the effect of being a recent immigrant was strong when controlling for self-employment status, in the final model for Ontario where all factors were accounted for, there was no significant effect of being an immigrant (Table A.12, Column 7).

The pattern for less-recent immigrants (i.e. those who had been in Canada between 10 and 29 years) was slightly different. There was a significant relationship with field of study as well as with major activity and industry / occupation (Table A.12, Column 3), signifying that if a less-recent immigrant was working and in the same industry / occupation as a Canadian worker, he or she would be more likely to fall into low earnings. This effect disappeared, however, once controlling for the influences of all other factors. Similar patterns were observed at the Canada level, though it should be noted that the effects were slightly stronger.

Nationally, provincial effects were somewhat weaker for the postsecondary-educated population than was the case for the total population aged 25 to 64. While postsecondary graduates with non-zero earnings in 2006 in Newfoundland and Labrador, Nova Scotia and New Brunswick continued to have higher odds of falling into the lowest earnings category, this was no longer the case for Prince Edward Island and Quebec. Moreover, there was no longer an Alberta advantage for postsecondary graduates compared to those living in Ontario.

### Education factors

The earnings advantage of having a university education compared to having a college education was greater in Ontario than it was at the Canada level. In Ontario, the odds of falling into the lowest earnings category stood at 0.53 for university-educated individuals compared to their college-educated counterparts, once all other factors were taken into account (Table A.12, Column 7). This effect held for all of the models and became slightly weaker only once industry and occupation were accounted for. This gap between college- and university-educated individuals was smaller at the Canada level, with an odds ratio of 0.67 for university graduates compared to those with a college education (Table B.12, Column 8).

The analysis finds that, once controlling for the influence of all other factors, no one field of study stands out as being more likely than the others to lead to an individual being in a low earnings situation. At the Canada level, the only field of study that was significantly different from the reference category was leisure studies, with an odds ratio of 2.39 (Table B.12, Column 8).

### Labour market factors

At the Canada level for the postsecondary-educated population, as was the case for the entire population aged 25 to 64, a worker's occupation was related to the probability of falling into low earnings. Workers in occupations in management, business administration, science, health, or the social sciences were all less likely to fall into low earnings than their counterparts working in sales occupations (the reference category) (Table B.12, Column 8). These effects continued to be significant even when other labour market characteristics were taken into account. In the case of Ontario, only occupations in management, science, and health were significantly less likely to be in a low earnings situation than those working in sales occupations (Table A.12, Column 7).

Certain industries were also associated with lower probabilities of falling into the lowest earnings category compared with the health industry (the reference category). Again, there are some differences between the findings at the Ontario and Canada levels. In the full models, at the Canada level, individuals working in primary industries, the finance industry, the professional, scientific and technical services industry and public administration were less likely than the reference group (the health industry) to fall into low earnings (Table A.12, Column 8). In contrast, individuals working in the food industry had significantly higher odds of being in the lowest earnings category. Similarly, in Ontario, the odds of falling into the lowest earnings category were significantly lower for individuals working in the finance industry, the professional, scientific and technical services industry and public administration, compared to the reference group (the health industry), whereas the odds were significantly higher for those working in the food industry (Table A.12, Column 7).

The final set of labour market factors to be discussed consists of those relating to main activity, work schedule and self-employment status. These factors have the largest impact on the likelihood that a postsecondary graduate (either college or university) would be in the lowest earnings group in both Ontario and at the Canada level in 2006.

In Ontario, the odds of being in the lowest earnings category compared to the reference category (working was the main activity for the year) was highest for students (odds ratio of 8.61), followed by retirees (odds ratio of 8.12) and finally, by those who reported that their main activity for the year was caring for children / a family member (odds ratio of 5.96) (Table A.12, Column 7). The overall pattern was similar at the Canada level for students (odds ratio of 8.63), but the odds ratios were lower for retirees (odds ratio of 5.33) and for those caring for children or a family member (odds ratio of 4.66) (Table B.12, Column 8).

Not surprisingly, in the cases of both Ontario and Canada, those who reported that they did not work full time, full year also were several times more likely than those working full time for the year to fall into the lowest earnings category. Finally, being self-employed was also associated with much higher odds of low earnings.3

## 6.3 The working population – multivariate analysis

Reporting an activity other than working as the major activity for the year accounted for a significant amount of the explanatory power of the two previous analyses. In order to fully understand the factors behind low employment earnings, it is necessary to focus the analysis on the working population only. Therefore, the third and final set of logistic regression analyses considers 1 (only those workers who reported that working was their major activity for the year and 2) postsecondary graduates who reported that working was their major activity for the year.

### Demographic factors

For the total working population, the gender effect again emerges as a statistically significant factor. It was weakest when work schedule was accounted for, indicating that more women working part-time helps to explain a part of the gender effect, and strongest when controlling for self-employment status, the effect being smaller in Ontario compared to the Canada level. In Ontario, after controlling for the influence of all factors, males had significantly lower odds of being in the lowest earnings category compared to females (odds ratio of 0.60; Table A.13, Column 6); at the Canada level, the odds ratio for men was even lower, 0.46 compared to women (Table B.13, Column 7).

When considering only postsecondary graduates in the working population, the gender effect retains its significance when demographic, education and field of study are taken into account, but is no longer significant in Ontario both in the model that controls for the effect of work schedule (Table A.14, Column 4) and in the final model that takes all factors into account (Table A.14, Column 7). This is not the case at the Canada level, however, where females have significantly higher odds of being in the lowest earnings category compared to males in all models, including the final model that takes all factors into account (odds ratio of 0.54 for men compared to women; Table B.14, Column 7).

The age effect was no longer significant for the working population in Ontario, nor was it for the working population of postsecondary graduates, with no age group being significantly more likely to be in a low earnings situation than another, once all factors were taken into account. While fairly strong age effects for workers aged 60 to 64 were observed in the preliminary models (Table A.14, Columns 1 to 5) these were no longer significant once self-employment status was accounted for, suggesting that at least part of the increased risk of older workers falling into lower earnings could be explained by their being self-employed. Thus, in the final model for Ontario, no age group in the postsecondary-educated working population was more likely to fall into low earnings than the reference category.

This was not the case at the Canada level where workers aged 55 to 59 had odds ratios of 1.33 compared to the reference group (45 to 49 year-olds) and workers aged 60 to 64 had odds ratios of 1.60 (Table B.13, Column 8). For the sub-group of postsecondary graduates in the working population at the Canada level, 60 to 64 year-olds had significantly higher odds of being in the lowest earnings category (odds ratio = 2.18; Table B.14, Column 8).

Differences between Ontario and the Canada level are also apparent when immigrant status is considered. In the interim models for the total working population in Ontario (Table A.13, Columns 1 to 5) there was no effect of being a recent immigrant. In the final model, however, recent immigrants (in Canada less than 10 years) showed higher odds of being in the lowest earnings category compared to non-immigrants (odds ratio of 2.10; Table A.13, Column 6). At the Canada level, immigrant status was highly significant in the total working population model, with odds ratios of 1.93 for recent immigrants and 1.49 for immigrants who had been in Canada for 10 to 29 years (Table B.13, Column 8). However, when considering only the postsecondary-educated working population, the immigrant effect was no longer significant at either the Ontario (Table A.14, Column 7) or the Canada (Table B.14, Column 7) levels.

Among the total working population at the Canada level, workers in Newfoundland and Labrador, Nova Scotia, New Brunswick and Quebec were all more likely to fall into low earnings compared to Ontario, whereas workers in Alberta were significantly less likely to do so (Table B.13, Column 8). In the case of the postsecondary-educated population, workers in Newfoundland and Labrador and Labrador, Nova Scotia and New Brunswick remained at higher risk of falling into the lowest earnings category compared to postsecondary-educated workers in Ontario. However, the odds ratios for postsecondary-educated workers in Quebec and Alberta were no longer significantly different from those of Ontario (Table B.14, Column 7).

### Education factors

For the entire working population, education level continued to have strong effects. At the Canada level, those with high school or less were more likely to fall into low earnings than high school or trade graduates, whereas college graduates were less likely to do so and university graduates even less so (Table B.13, Column 8). This was not the case in Ontario, however, where the only statistically significant finding was that university graduates had much lower odds of being in the lowest earnings category compared to other workers (odds ratio of 0.48; Table A.13, Column 6).

The relative advantage of university graduates over college graduates holds within the postsecondary working population as well. In Ontario, the odds ratio for university graduates compared to college graduates was 0.51 (Table A.14, Column 7) and at the Canada level, it was 0.66 (Table B.14, Column 7).

### Labour market factors

Finally, and not surprisingly, working time played a major role in determining who was more likely to be in a low earnings situation, with individuals who worked on less than a full-time, full-year basis being far more likely than full-time, full-year workers to have earnings that placed them below half the national median employment earnings. Self-employed workers were also more likely than those working for an employer to report low earnings.

## Notes

1. In order to be consistent with the data for Canada reported by the Organization for Economic Cooperation and Development (OECD), individuals with non-zero employment earnings in 2006 were retained in the analysis.
2. The Canada-level models are slightly different than the Ontario models in that they include a control for province effects.
3. Recall that the self-employed present a special case in that, while they may have non-zero earnings from employment, their earnings are treated differently in the tax system, with earnings being lower as a result of the deduction from income of employment-related expenses. Workers in an employer-employee relationship generally cannot deduct such expenses from their employment earnings.
Date modified: