This paper assesses the contribution of cohort effects, age composition effects, and macroeconomic factors in the evolution of the female labour force participation rate in Canada between 1976 and 1994. Using data from the Survey of Consumer Finances, we find that cohort effects are the main factor behind the recent stagnation in female participation rates. Though the poor macroeconomic performance of the Canadian economy during the 1990s has also contributed to this phenomena, it cannot explain in itself why the behaviour of female participation rates in the 1990s was so different than in previous decades. We reach similar conclusions when we analyse the evolution of the employment rate.

One related finding is that both the level and the slope of the age-participation profiles of women have changed over time. While older cohorts had profiles that were sloping up between the ages of 25 and 50, younger cohorts exhibit much flatter (and higher) profiles for the same age range. In other words, age-participation profiles of women increasing look like those of men which are flat at very high levels before declining after age 50.

Dans ce texte, nous évaluons le rôle des effets de cohortes, de la structure d'âge et des facteurs macro-économiques dans l'évolution du taux d'activité des femmes au Canada entre 1976 et 1994. Le résultat principal qui ressort de notre analyse des données de l'enquête des finances des consommateurs est que les effets de cohortes sont le facteur clef permettant d'expliquer le plafonnement récent de ces taux d'activité. Bien que la performance macro-économique peu enviable de l'économie canadienne dans les années quatre-vingt-dix ait elle aussi contribué à ce phénomène de plafonnement, il n'en reste pas moins que seuls les effets de cohortes parviennent à expliquer pourquoi les taux d'activité ont crû beaucoup plus rapidement dans les années soixante-dix et quatre-vingt que dans les années quatre-vingt-dix. Nous tirons les mêmes conclusions lorsque nous analysons les taux d'emploi plutôt que les taux d'activité.

Il ressort aussi de notre étude qu'à la fois le niveau et la pente des profils de participation en fonction de l'âge (profils d'âge) ont changé à travers le temps. Alors que la pente des profils d'âge était positive entre l'âge de 25 et 50 ans pour les cohortes entrées depuis longtemps sur le marché du travail, ces profils sont beaucoup plus plats pour celle entrées plus récemment sur le marché. En d'autres termes, les profils d'âge des femmes ressemblent de plus en plus à ceux des hommes qui sont, eux aussi, plutôt plats jusqu'à l'âge de 50 ans.

Introduction

This paper studies the behavior of labour force participation among women between the ages of 25 and 64 for the years 1976 to 1994. The goal of this study is to determine whether the stagnation, and in some cases decline, in the female labour force participation rate in the 1990s is a temporary phenomenon tied to the poor growth performance of the economy or whether it is an ongoing effect, which may signal that the process of integration of women into the work force is almost complete. A cohort analysis is used to examine this issue.

The methodology involves isolating the effect of three separate factors on the participation rate of women. To this end, we follow cohorts of women over time, that is, we track the participation rate of representative groups of women who entered into the work force at a given point in time (e.g. women who were 25 years old in 1976). We then decompose a cohort's participation rate into, first, a macroeconomic effect that by definition is common across cohorts. Recession and structural phenomena such as the generosity of the employment insurance system are some of the factors that may cause a macroeconomic effect. The second factor is the age or life cycle effect, which shows how the cohort's participation rate changes as the cohort ages. The third factor is the cohort specific effect, which shows differences between cohorts for a given age and macroeconomic effect. For example, if the cohort that entered the labour force in 1976 has a participation rate that is 10 percent higher than that of the cohort that entered the labour force in 1966 at the same age and under similar macroeconomic conditions, the 1976 cohort is said to exhibit a 10 percent cohort effect relative to the 1966 cohort.

Our results indicate that cohort effects are likely the dominant factor in explaining the recent stagnation in female labour force participation rates. The same result was obtained when the labour force participation rate was replaced by the employment rate. Cohort effects help explain both the large increase in participation and employment rates during the 1970s and 1980s, as well as their stagnation in the 1990s. The 1989-1994 recession merely amplified the stagnation phenomenon; it also explains the observed decline in a number of demographic groups. These results show, however, that stagnation would have occurred even had more favourable macroeconomic conditions prevailed.

This paper is divided into five sections. Section 1 outlines the data used and provides some illustrative graphs. Section 2 presents the actual cohort analyses. In Section 3, the results of those analyses are used to break down the evolution of the labour force participation rate into macroeconomic, age and cohort effects. Section 4 examines result robustness, studying both the influence of changes in the population's education level on changes in the participation and employment rates, and the effect of the generosity of the employment insurance program, a major structural macroeconomic factor. And finally, Section 5 makes some projections for future participation and employment rates. Participation and employment rates will be analysed simultaneously to ascertain that the evolution of the participation rate does not merely reflect changes in the way Canadian women "classify" themselves in the labour market.

Section 1: Data and descriptive statistics

The data used were obtained from the Survey of Consumer Finances (SCF) for the years 1976, 1978, 1980, 1982, 1983, 1985, 1987, 1989, 1991, 1993 and 1994 (survey years). These years were used because: 1) from 1976 to 1982, the survey was taken only every other year; 2) since then, the survey has been taken every year except 1984; and 3) 1994 was the last year available when we began this study. The years represented (about every other year) provide a fairly coherent sample across time.

Individuals were grouped into two-year cohorts according to their date of permanent entry into the labour force. This was defined, somewhat arbitrarily, as the even-numbered year in which the woman in question was 25 or 26 years old (e.g. a woman born in 1941 is in the "entering" cohort for 1966). Thus for each even year, all women between 25 and 64 are divided into 20 cohorts (25-26, 27-28, ..., 63-64). In total, 29 cohorts entered into the labour force between 1936 and 1992.

Note that Statistics Canada's public use files of the SCF from before 1982 provide data only for heads of households and spouses; we therefore confined our analysis to this sub-sample for the entire 1976-1994 period. Labour force activity (employment, unemployment or non-participation) is determined based on individual responses to the usual LFS questions (for the month of April in the SCF). The evolution of the labour force participation rate, represented by a solid line, and the employment rate, represented by a dotted line, for each cohort is shown in Figure 1 (age is indicated in each graph). All the information used in this study is presented in a raw form in this figure. Those cohorts that entered the labour force first are shown only in the early years, while those that entered last appear only in later years. Only the "middle" cohorts (those that entered the labour force between 1954 and 1972) are shown in all years.

The figure shows that the evolution of the labour force participation and employment rates is similar for all cohorts. Both these rates tend to increase from the age of 25 to 45-50 years, then decrease rapidly until age 65. Participation and employment rates are obviously higher for those cohorts that entered the labour force most recently than for the others.

Aggregate data (ages 25-64, 25-44 and 45-64) are presented in Figure 2. The figure reproduce for our SCF sample the trends that we are trying to explain, i.e. the stagnation, and in some cases decline, in participation and employment rates for all age brackets but the oldest (45-64 years) in the female population.

Other descriptive statistics are presented in Table 1, which illustrates age composition and education levels (percentage of women with a high school education or less) for each of the years studied. The table shows quite a young population during the period from 1976 to 1994. About 65 percent of women between the ages of 25 and 64 during these years were 44 or younger. The impact of the baby boom/baby bust on the population's age composition is also clearly visible. This helps explain the increase in the proportion of women aged 35-44 since the beginning of the 1980s; the first wave of boomers born in 1946 reached the age of 35 in 1981. The same phenomenon occurred at the beginning of the 1990s as the first of the boomers reached 45. It is now the baby bust generation, those women born after 1965, who make up the 25-34 year-old segment.

The statistics presented in Table 1 also show a steady increase in level of education: the percentage of women with a high school education or less dropped from 73.6 percent in 1976 to 54.3 percent in 1994. This trend, however, is slightly exaggerated by the revamping of the questions on education in the LFS in 1990.

Section 2: Cohort Analysis

2a. Econometric Model

An econometric model is used to examine the separate roles played by the macroeconomic, cohort and age effects on labour force participation and employment rates. The dependent variable used in the regressions is the participation (or employment) rate p_jt for cohort j at time t expressed in "log-odds" form ln(p_jt/(l-p_jt)). For example, p_74,84 represents the labour force participation rate for the cohort that entered the labour force in 1974 (j = 74) during the year 1984 (t = 84). This functional form is used to account for the special nature of variable p_jt, whose value is always between 0 and 1. It ensures that the predicted value will always be between 0 and 1, which would not be the case if a standard linear specification were used instead.

In most of the estimated models, only one macroeconomic variable is used, the unemployment rate among men aged 25 to 44. Although certain long-term trends in this rate may be determined by structural factors, it is clear that its short-term fluctuations are mainly reflective of the evolution of the economic climate. Other variables such as the output gap may be used in addition to unemployment rate, but we prefer to concentrate on the latter, because of its simplicity; however, the results must be interpreted with caution. The scope of the macroeconomic effect will, however, be broadened in Section 4 by adding other variables.

For the age effect, a fourth degree polynomial that allows variations in participation rate over the entire life cycle is used. A flexible functional form is also used to show the cohort effect. A third degree polynomial yields the equation:

One characteristic of equation (1) is that the age profile for each cohort, i.e. the evolution of the labour force participation rate over the life cycle, is similar for each cohort; they differ only in terms of the intercept. In other words, the model allows a vertical displacement of the life cycle profile from one cohort to another while forcing the shape of the profile, and thus the slope, to be identical for each cohort. A more general model is produced by introducing age-cohort interaction terms to allow the age effect to vary from one cohort to another. This was done with the following model, which incorporates an age-cohort (a_jtj) and an age-cohort squared interaction term:

(2) ln(p_jt/(1-p_jt)) = α +δur_t +β₁j +β₂j² +γ₁a_jt +γ₂a_jt² +γ₃a_jt³ + γ₄a_jt⁴ +θ₁a_jtj +θ₂a_jtj²

If second or higher order polynomial terms are omitted, equation (2) shows that the age effect on ln(p_jt/(1-p_jt)) is equal to γ₁+θ₁j. If θ₁ is positive, the age effect will be greater for those cohorts which most recently entered the labour force (highest j) than for the others, and vice versa. Coefficient θ₁ thus allows the life cycle profile to vary from one cohort to another.

Graph 1 illustrates the advantages of equation (2) over equation (1), which does not include the age-cohort interaction term. Without such interaction terms, the intercept is the only difference between different cohorts’ age profiles (Graph 1a). The same increase in participation at career outset and the same decrease in participation at career end is shown for every cohort. The age profile is clearly more flexible in Graph 1b where interaction terms are introduced. In this graph, the "new" cohort has both a higher ordinate value at the origin and a shallower slope. This results in a higher and more stable age profile at career outset than in the previous cohorts (the "old" cohorts on the graph). This profile is also more similar to that for men, whose participation rates are fairly high and stable until the age of about fifty. The situation shown in Graph 1b is therefore more consistent with the idea of a convergence between men’s and women’s labour force participation rates, or increasing participation of women in the labour force, than that shown in Graph 1a.

In Graph 1b, the cohort effect is concentrated at career outset, participation rates before the age of 40 for the new cohort being much higher than those for the old cohort, while the rates are reasonably comparable after the age of 50. The impact of the entry of the new cohort on the aggregate labour participation rate would thus be felt most strongly during the first 10 or 20 years after the cohort’s arrival, while in Graph 1a, its influence is shown as continuing throughout the life cycle. In other words, the entry of new cohorts in Graph 1b should result in a rapid increase in the aggregate participation rate, followed by a period of stagnation. Graph 1a, on the other hand, shows a constant increase in the aggregate participation rate.

Graph 2 illustrates the impact of the arrival of new cohorts in the two cases discussed above, with those cohorts entering the labour force after 1970 considered "new" cohorts and those entering before 1970 considered "old" cohorts. The graph clearly shows that only the presence of an age-cohort interaction effect explains the stagnation phenomenon.

An often-mentioned problem with cohort analyses is the impossibility of separately identifying cohort effects, year effects (macroeconomic effects), and age effects because of the linear dependence between them. In fact, since a_jt=25+t-j, the three variables (a_jt, j and t) are perfectly collinear. This study proceeds on the implicit assumption that variable ur_t captures any systematic macroeconomic effect and that there is no other temporal trend in this effect. That said, econometric models (such as (1) and (2)) can never explain all the variations in the data (R　squared is less than 1). As a rule, a residual macroeconomic effect is obtained, representing the macroeconomic variation in the data that cannot be explained by other variables in the model. If during a period, say the nineties, we were to find a large residual we would interpret this as indicating that participation in this period has experienced a macroeconomic effect not captured by its standard comovement with the unemployment rate.

2b. Results

Equations (1) and (2) were estimated using weighted ordinary least squares, with cohort size j at time t used as the weights. The results are shown in Table 2 for employment rates (columns 1 to 3) and participation rates (columns 4 to 6). For model (1), note that all the coefficients are significant except for the rate of unemployment among men 25 to 44 (columns 1 and 4). That effect becomes significant, however, when age-cohort interaction terms from model (2) are introduced (columns 2 and 5). Also note that interaction term coefficients are highly significant, and that R squared for model (2) is higher than for model (1).

We also present the results of the regressions when the sample is limited to the 1976-1989 period (columns 3 and 6). The purpose of this exercise is to assess whether the levelling off of participation and employment rates in the 1990s was predictable from the behaviour of these rates prior to 1990. The results indicate that the estimated parameters for 1976-1989 are very similar to those for the period as a whole. We shall return to the question of the stagnation of the participation and employment rates.

To facilitate presentation of the results, it is simpler to use a graphical approach than to examine the numbers presented in Table 2 in detail. For each rate (participation and employment) and each model (1 and 2), we present the following four graphs: Graph (a) shows the cohort effect at age 44 — i.e. the variations in the participation and employment rates attributable to the cohort effect at a precise point in the life cycle. Graph (b) shows the age effect throughout the life cycle for a typical cohort (the one which entered the labour force in 1964). Graph (c) presents a similar result for six cohorts (the ones which entered the labour force in 1940, 1950, 1960, 1970, 1980 and 1990) to illustrate cohort differences over the entire age profile. It should be noted however that Graph (b) shows a predicted age profile for the entire life cycle while Graph (c) shows the profile only for the ages at which the cohort in question is observed in the data (1976 to 1994). Finally, Graph (d) indicates the degree to which the estimated model for 1976-1989 can be used to predict participation and employment rates for the entire 1976-1994 period.

Rather than discussing each of the graphs in detail, we will confine ourselves to noting a few highlights:

Our results seem to indicate that in addition to unfavourable macroeconomic conditions, the levelling off of cohort effects also contributed to the trend observed in the 1990s. This hypothesis will be examined in detail in the following section.

Section 3: Decomposition

We shall now perform a more formal analysis of the role of different factors in the recent evolution of aggregate participation and employment rates for all women aged 25-64 by decomposing this evolution into four components: the macroeconomic effect related to the unemployment rate among men aged 25-44 (the economic cycle), the residual macroeconomic effect, the age effect and cohort effects. In terms of equation (2), it is relatively easy to identify the first two factors, which correspond to the term δur_t and to the residuals of this equation.

More precisely, we first calculate the participation (or employment) rate for each year, taking the weighted average of p_jt values for each t. The observed rate (p_jt) is then replaced by the predicted rate _jt from the estimated model. The average of the _jt values for each t therefore represents the aggregate rate predicted by the model. The difference between the observed aggregate rate and the predicted rate represents the residual macroeconomic effect.

We then recalculate the prediction by replacing the observed unemployment rate by the average of the unemployment rates over the entire sample (8.2 %). The difference between this new prediction and the preceding prediction represents the macroeconomic effect related to the male unemployment rate, which we also call the cyclical effect.

The cyclical and residual effects obtained in this manner are presented in Figure 7a for the participation rate and 8a for the employment rate. During the 1990s, the cyclical effect is in the order of -1% for the participation rate and -2% for the employment rate. In other words, the female employment rate would have been 2% higher in the 1990s if the male unemployment rate had held steady at 8.2%.

Age and cohort effects are somewhat more complicated to understand because of the interaction terms in equation (2). It should be noted, first of all, that the age effect comes into play only to the extent that the population's age composition changes over time. For example, the arrival of the baby boomers in the labour force in the early 1970s considerably rejuvenated the 25-64 year-old population as a whole. As these young women had below-average participation rates, it should have been expected that this change in composition would have had a negative effect on the aggregate participation rate and vice versa.

It might therefore be supposed that to identify the age effect, it is enough to recalculate the predicted rate using a uniform age composition (5% of the population aged 25-64 in each 2-year age group) instead of the observed age composition. The problem with this procedure is that it depends on the cohorts present in the labour force in each year, since the age profile is dependent on the cohorts through the interaction terms. This procedure therefore serves to isolate the age effect plus the crossed age-cohort effect.

The same problem arises when we want to isolate the role of cohorts. For example, we can try to recalculate the predicted rates by replacing the cohort effect expressed as β₁j +β₂j² +θ₁a_jtj +θ₂a_jtj² by the cohort effect obtained if the cohort is set at an arbitrary level such as j=70 (β₁70 +β₂70² +θ₁a_jt70 +θ₂a_jt70²). This gives us the cohort effect plus the crossed age-cohort effect, in the same way as in the case of age. Once we have all this information, however, it is possible to calculate the "pure" age effect (for a given cohort composition) and the joint age-cohort effect separately.

These different effects are illustrated in figures 7b, 7c and 7d for the participation rate and figures 8b, 8c and 8d for the employment rate. Let us take the example of Figure 7b: in this case, we use the cohort which entered the labour force in 1970 as a reference cohort for the decomposition. The cohort effect thus indicates the difference between the observed rates and the rates which would have prevailed had all the cohorts followed the same age profile as cohort 17, other factors being kept constant. This cohort effect is therefore the "pure" effect mentioned earlier. The graph also shows the "pure" age effect (for a given cohort and other factors) as well as the combined age and age-cohort effect (the age effect for the observed cohorts in each year).

While it can be rather difficult to grasp all the details of these decompositions, the results speak for themselves: it is really the cohort effect that dominates the evolution of the participation and employment rates throughout the 1976-1994 period. The results are very similar regardless of which cohort is used as a reference for the decompositions (1970, 1980 or 1990). We find that cohort effects account for an increase of about 20 percentage points in participation and employment rates between 1976 and 1994. At the same time, the graphs clearly indicate that this phenomenon seems to be coming to an end. By comparison, age effects play a relatively small role in recent changes.

To sum up, our results indicate that the stagnation of female participation and employment rates is primarily a structural phenomenon related to the stabilization of the cohort effects which were responsible for the remarkable increase in these rates in the 1970s and 1980s. The unfavourable macroeconomic situation amplified this phenomenon but was not the root cause. The relative performance of the participation and employment rates during the 1981-1983 and 1989-1994 recessions clearly illustrates this phenomenon; in 1981-1983, the downward pressure on the rates from the macroeconomic effect was offset by the cohort effects, pushing the rates up by one percentage point per year, whereas in 1989-1994, due to the stabilization of cohort effects, macroeconomic effects comparable to those of 1981-1983 resulted in lower participation and employment rates.

To clarify the role of cohort effects, we illustrate their magnitude at age 24, 34, 44, 54 and 64 in figures 9 and 10. Let us take for example Figure 9c, which shows the cohort effect at age 44 by year of entry into the labour force. The vertical line indicates the cohort which was aged 44 in 1994. The curve to the left of the line describes the evolution of cohort effects during the 1976-1994 period. The curve to the right shows the predicted evolution for the coming years (see Section 5 for more details).

The results indicate a general slowing trend for most of the ages under consideration, attributable to cohort effects. This is particularly true for the younger groups (ages 24 and 34), which explains why the levelling off and declining trend is more pronounced for the 25-44 age bracket than for the 45-64 bracket (figures 2a and 2b).

Section 4: Robustness analysis

4a. Education

We re-estimated the models separately for women who have pursued post-secondary studies and those who have only a high school diploma or less. The highlights of the results presented in figures A3, A4, A5 and A6 are:

In Table 3, we present the results of the regressions when the employment insurance subsidy rate is also used as a macroeconomic variable. The results are not very conclusive, since the effect on the employment rate is negative when we also control for the unemployment rate among men aged 25-44 (column 2). The decline in the subsidy rate during the 1980s should therefore have increased the employment rate instead of lowering it. The effect on the participation rate is not significant (column 4). This being said, including the subsidy rate as a macroeconomic variable has little impact on the model's other coefficients. Our conclusions about the role of cohort effects versus macroeconomic effects during the 1990s therefore remain unchanged.

Section 5: Predictions

We will now attempt to predict the future evolution of the participation and employment rates under two different macroeconomic scenarios: a 8.2% unemployment rate for men aged 25-44 (the average over the 1976-1989 period) and a 6.6% unemployment rate for the same group (the 1989 level). To do so, we must make some assumptions about the cohorts which will enter the labour force after 1994. To simplify the exercise, we will simply hypothesize that the size and age profile of these cohorts will be similar to those of the last cohort observed (the one which entered the labour force in 1992).

The results of the simulations are presented in figures 11 (participation rate) and 12 (employment rate). The conclusions are the same in both cases: large increases in the participation and employment rates are clearly a thing of the past; in the future, these rates can be expected to hold relatively stable. However, there is still room for a 2-3 percentage point increase in the rates if the macroeconomic situation continues to improve. It is illusory, though, to think that the rates could rise 5-10 percentage points during the next period of expansion as they did over the 1983-1989 period. The cohort effects which prevailed at the time are simply no longer present today.

Conclusion

This study's main finding is that the levelling off of female participation and employment rates is primarily a structural phenomenon related to the stabilization of the cohort effects which accounted for the remarkable increase in these rates in the 1970s and 1980s. The unfavourable macroeconomic situation has amplified this phenomenon but is not the root cause. The relative performance of the participation and employment rates during the 1981-1983 and 1989-1994 recessions clearly illustrates this phenomenon; in 1981-1983, the downward pressure on the rates from the macroeconomic effect was offset by the cohort effects, pushing the rates up by one percentage point per year, whereas in 1989-1994, due to the stabilization of cohort effects, macroeconomic effects comparable to those of 1981-1983 resulted in lower participation and employment rates.

This result is strongly dependent on the amount of flexibility used to capture cohort effects. It is essential that the age profile as a whole, and particularly its slope, be allowed to vary from one cohort to another. This makes it possible to accurately trace both the rise and the flattening of the employment and participation profiles by age. These phenomena are consistent with a convergence in the behaviour of men and women in the labour market: men exhibit very high and very flat (at least until age 55) employment and participation profiles over their life cycle. The profiles of recent female cohorts are therefore closer to those of men than to those of older female cohorts.

Finally, the recent evolution of participation and employment rates in the U.S. seems to corroborate our findings: a levelling off of the employment and participation rates has been observed there as well, despite more favourable macroeconomic conditions in the U.S. than in Canada since 1992. There too, the structural phenomenon of cohort effects seems to be the dominant factor in the evolution of female employment and participation rates since the 1970s.

Beaudry, Paul, and David Green, "Cohort Patterns in Canadian Earnings: Assessing the Role of Skill Premia in Inequality Trends," National Bureau of Economic Research Working Paper No. 6132, August 1997.

Card, David, and W. Craig Riddell, "A Comparative Analysis of Unemployment in Canada and the United States," in D. Card et R. Freeman (eds.) Small Differences that Matter: Labor Markets and Income Maintenance in Canada and the United States, Chicago: University of Chicago Press for NBER, 1993, pp. 149-189

Table 1: Descriptive statistics

Year	Age distribution in %				% with high school or less	Number of obs.
	25-34	35-44	45-54	55-64
.	(1)	(2)	(3)	(4)	(5)	(6)
1976	40.9	24	19.8	15.3	73.6	12269
1978	39.1	24.9	20.5	15.5	75	17372
1980	39.9	26.1	19.1	14.9	74.7	18212
1982	39.5	26.2	19.1	15.3	73.5	18881
1983	39.2	27.5	18.9	14.4	72.7	19775
1985	38.7	29.3	17.7	14.3	71	19664
1987	37.9	29.9	18	14.3	70.4	17949
1989	37.1	31	18.3	13.6	68.3	21117
1991	34.9	32.3	19.3	13.6	59.7	26033
1993	31.4	32.8	21.6	14.2	56.2	22592
1994	32.8	32.7	21.2	13.3	54.3	22420
Total	37	29.2	19.4	14.3	67.2	216284

Table 2: Detailed results of regressions
(standard deviation in brackets)

.	Employment rate			Participation rate
	1976-1994	1976-1994	1976-1989	1976-1994	1976-1994	1976-1989
	(1)	(2)	(3)	(4)	(5)	(6)
Constant	0.792 (0.143)	-3.102 (0.291)	-2.912 -0.344)	0.997 (0.158)	-3.398 (0.245)	-2.746 (0.177)
Unemploy. rate	-0.019 (0.052)	-0.029 (0.004)	-0.031 (0.003)	-0.004 (0.011)	-0.016 (0.003)	-0.014 (0.002)
Cohort effect:a	.
co	-1.83 (0.154)	3.804 (0.383)	3.506 (0.465)	-2.227 (0.151)	4.162 (0.319)	3.267 (0.215)
co^2	1.609 (0.131)	-0.792 (0.118)	-0.681 (0.15)	1.892 (0.102)	-0.851 (0.1)	-0.561 (0.062)
co^3	-0.289 (0.025)	.	.	-0.336 (0.019)	.	.
Age effect:	.
Age/10	0.389 (0.053)	2.706 (0.175)	2.78 (0.216)	0.405 (0.051)	3.004 (0.124)	2.749 (0.126)
(Age/10)^2	-0.445 (0.024)	-1.742 (0.101)	-0.676 (0.027)	-0.471 (0.021)	-0.76 (0.032)	-0.701 (0.019)
(Age/10)^3	-0.13 (0.007)	-0.151 (0.007)	-0.161 (0.009)	-0.139 (0.006)	-0.162 (0.006)	-0.167 (0.009)
(Age/10)^4	0.031 (0.006)	0.032 (0.006)	0.024 (0.004)	0.032 (0.005)	0.034 (0.005)	0.027 (0.005)
Interactions:	.
Age/10 *co	.	-1.742 (0.101)	-1.902 (0.14)		-1.963 (0.067	-1.886 (0.091)
(Age/10) *co^2		0.226 (0.021)	0.298 (0.021)		0.258 (0.016	0.305 (0.021)
R squared:	0.941	0.954	0.951	0.953	0.964	0.962
Number of observations	224	224	164	224	224	164

Table 3: Unemployment insurance effect
(Standard deviation in brackets)

.	Employment rate		Participation rate
	(1)	(2)	(3)	(4)
Constant	-2.654 (0.501)	-2.984 (0.294)	-3.157 (0.361)	-3.337 (0.249)
Unemploy. rate		-0.023 (0.006)		-0.012 (0.006)
UI subsidy rate	-0.117 (0.03)	-0.074 (0.037)	-0.062 (0.029)	-0.038 (0.04)
Cohort effect:	.
co	3.486 (0.682)	3.858 (0.401)	3.987 (0.51)	4.189 (0.35)
co^2	-0.778 (0.209)	-0.841 (0.128)	-0.841 (0.159)	-0.876 (0.116)
Age effect:	.
Age/10	2.546 (0.331)	2.733 (0.181)	2.916 (0.232)	3.018 (0.135)
(Age/10)^2	-0.7 (0.063)	-0.715 (0.046)	-0.757 (0.049)	-0.766 (0.039)
(Age/10)^3	-0.151 (0.007)	-0.151 (0.007)	-0.162 (0.006)	-0.163 (0.006)
(Age/10)^4	0.032 (0.006)	0.032 (0.006)	0.033 (0.005)	0.033 (0.005)
Interactions:	.
Age/10 *co	-1.727 (0.197)	-1.791 (0.109)	3.987 (0.51)	-1.988 (0.084)
(Age/10) *co^2	0.226 (0.021)	0.226 (0.021)	-0.842 (0.159)	0.258 (0.016)
Rsquared	0.952	0.955	0.964	0.964
Number of observations	224	224	224	224