Notes

^1.

Human Resources Development Canada (2000) also recognizes that the self-employed can be divided into two groups of individuals: individuals whose aim is to earn a "decent" wage, not necessarily to grow their businesses; and individuals who want to expand their businesses. This report also identifies the own-account self-employed with the former group and employers with the latter group.

^2.

See Leung (2006), for example.

^3.

See Schuetze (2005) for a review of the literature in this area.

^4.

Kuhn and Schuetze (2001) also used non-employment as the base category. Choosing non-employment as the base category makes the estimated coefficients easier to interpret because key variables such as employment in the previous period, education, and age all increase the returns to paid employment and self-employment relative to non-employment. That is to say, examining the difference in coefficients across time and employment states is easier when all coefficients have the same sign.

^5.

Prowse (2007) also showed the conditions where identification in a dynamic MML model does not depend solely on the functional form of the unobserved terms.

^6.

A possible alternative specific explanatory variable that is available in the data is earnings. In occupations in which the individual is not observed, this individual's earnings could be predicted by earnings regressions. Following papers such as those by Rees and Shah (1986) and Leung (2006), a reduced-form occupational choice model would be first estimated, and sample selection correction variables would then be created and used in the self-employment and wage-employment earnings regressions. Predicted earnings from these regressions could be used in the "structural" occupational choice model. This possibility is not pursued here because self-employment earnings are difficult to explain with conventional explanatory variables. The R² from self-employment earnings regressions are much lower than those from wage-employment earnings regressions. For example, see Hamilton (2000). The possible unreliability of reported self-employment earnings is one of the reasons why this paper looks at the labour market dynamics, not at earnings, to characterize the self-employed.

^7.

Esmaeilzadeh (2008) allows each state-specific random effect to follow a discrete multinomial distribution. Since the random effects are not allowed to be correlated, the approach in that paper is not necessarily less restrictive than the one taken in this paper.

^8.

Female self-employment rates are lower than male self-employment rates. This is particularly the case for self-employment with paid help. Over the 1993-2007 period, the Labour Force Survey shows that the rate of self-employment with paid help for females ranged from 2.8% to 3.2%, while the rate for males was 7.0% to 8.8%. The analysis of transitions exacerbates the problem of a smaller self-employment sample for females; consequently, this paper focuses solely on males. As in the case of males, it can be shown that female self-employment was more persistent in the 2000s than in the 1990s. These tables are available upon request.

^9.

The conclusions of the paper are not substantially affected by the exclusion of government and agricultural workers. Results including these workers are available upon request.

^10.

See, for example, Lin, Picot, and Yates (1999).

^11.

SLID does ask how long one has been non-employed, but it does not ask how long one has been in self-employment or in paid employment.

^12.

Only immigrants who arrived in Canada after the age of 18 are considered immigrants for purposes of this paper.

^13.

The non-employed are assigned the industry and occupation of their last job. If they were non-employed at the beginning of the panel, they are assigned the industry and occupation of their first job. This is possible because all individuals in the restricted sample had at least one job over the six-year period.

^14.

In past studies, means of the explanatory variables are sometimes also included in order to control for initial conditions, but there is so little time variation in the short panel that collinearity problems arose when these variables were added.

^15.

What is reported in Tables 4, 5, 6 and 7 are the results of a t-test; the difference in the coefficients between two time periods, divided by the square root of the sum of their variances.

^16.

The unemployment rate is also among the variables that are statistically significant. Higher unemployment lowers the returns to all employment states—paid employment, own-account self-employment, and self-employment with paid help—relative to non-employment. Poorer economic conditions make a move into non-employment more likely to increase the probability of staying longer in non-employment. Another interesting difference between the 1990s and the 2000s is that the unemployment rate is no longer statistically significant in the 2000s. This could be because there is less variation in the unemployment rate in the 2000s, or it could be that the effect of the unemployment rate depends on the nature of the unemployment. For example, Moore and Mueller (2002) suggest that the length of joblessness and the type of separations occurring are important factors.

^17.

Compared to the marginal effects of the lagged employment state variables, the other marginal effects are small, but as expected. Relative to having a high-school education, having a less than a high-school education increases the probability of being non-employed. Moving from a high-school education to a university degree raises the probability of being in paid employment; being an immigrant lowers the probability of paid employment and raises the probability of own-account self-employment; and being in manufacturing relative to trade lowers the probability of being self-employed and raises the probability of being in paid employment.

^18.

Experiments where smaller groups of coefficients and characteristics are allowed to change in turn do not yield more insight into the differences between the 1990s and the 2000s.

^19.

Data from the Labour Force Survey show that, between 2000 and 2007, 84% of the self-employed with less than one month's tenure were unincorporated. Over the same period, 61.6% of all self-employed were unincorporated.