outcomes within the cells are sample statistics, and a random effects estimation procedure should be used.
5. Regression results
Parameter estimates for the regression on labour force participation are reported in the first column of Table 3. Most of the control variables have the expected
signs and are significantly different than zero at conventional test levels. Maori and Pacific Islanders are less likely to participate in the labour force. Marital
status, and the number and ages of children in the family have substantially different effects on participation between men and women. Individuals outside the
prime working ages of 25–54 are less likely to be in the labour force. Participation increases with educational attainment. Although dummy variables representing all
41 quarters in this data set are included in the regressions, these results are not reported.
The effects of qualitative changes in welfare programmes on labour force participation are mixed. The increase in the age of eligibility for Superannuation
has a strong positive effect on participation. The estimated coefficient of 0.051 is significant at better than a 1 level. Independent of any associated effects of this
policy on either maximum weekly benefits or breakeven income levels, raising the age of eligibility for publicly funded retirement benefits has increased the labour
force participation of this age group.
Contrary to our original hypothesis, the increase in the age of eligibility from 16 to 18 under both UB and DPB reduced labour force participation. The
estimated coefficient of y0.095 is significant at better than a 1 level. Since these youth are potentially eligible for benefits under other programmes with much
tighter eligibility criteria but similar maximum weekly benefits, earnings disre- gards and benefit reduction rates, this dummy variable captures any change in
participation among 16- or 17-year-olds. It suggests that the LFPR of 16- or 17-year-olds declined by 9.5 percentage points as a result of this tightening in
eligibility criteria. We return to this issue later.
The estimated coefficient on the natural logarithm of the real maximum weekly benefit is y0.084. With a standard error of 0.009, this estimated coefficient is
statistically significant at better than a 1 level. This is consistent with our null hypothesis that a reduction in the guarantee increases aggregate labour supply. A
10 cut in potential benefits raises participation by 0.84 percentage points. This result can be converted into an elasticity by dividing by the estimated coefficient
Ž .
by the mean of the dependent variable 0.746 . This estimated labour supply elasticity is y0.113.
The estimated coefficient on the natural logarithm of the real breakeven income level is y0.200, and statistically significant at better than a 1 level. This says
that a reduction in breakeven income associated with either a fall in the real
Table 3 Estimated determinants of labour supply random effects estimation
Standard errors are in parentheses. Observations were weighted by population weights constructed by Statistics New Zealand. Dummy variables representing the 41 quarters in this data set were included in
these regressions, but these results are not reported. Random effects were based on the 1383 cells of individuals observed over the sample period.
Independent variables Dependent variables
Participation in the Weekly hours of
Participation in the labour force
labour supplied labour force or education
UU UU
UU
Ž .
Ž .
Ž .
Constant 2.396
0.069 71.431
4.215 3.101
0.051
UU UU
UU
Ž .
Ž .
Ž .
Maori y0.053
0.003 y2.788
0.232 y0.057
0.002
UU UU
UU
Ž .
Ž .
Ž .
Pacific Islander y0.045
0.004 y2.798
0.278 y0.022
0.003
UU UU
UU
Ž .
Ž .
Ž .
Married 0.213
0.005 10.070
0.369 0.245
0.003
U UU
Ž .
Ž .
Ž .
Child aged - 2 in family 0.020
0.010 y0.532 0.556
0.019 0.007
UU UU
Ž .
Ž .
Ž .
Child aged 2–4 in family y0.037
0.011 y0.680 0.574
y0.051 0.008
UU UU
UU
Ž .
Ž .
Ž .
Average number of 0.022
0.003 y0.495
0.178 0.048
0.002 children in family
UU UU
UU
Ž .
Ž .
Ž .
Female y0.081
0.004 y6.074
0.327 y0.069
0.002
UU UU
UU
Ž .
Ž .
Ž .
Female P married y0.064
0.005 y6.009
0.422 y0.085
0.003
UU UU
UU
Ž .
Ž .
Ž .
Female P child aged y0.356
0.012 y10.765
0.696 y0.408
0.010 -
2 in family
UU UU
UU
Ž .
Ž .
Ž .
Female P child aged y0.137
0.014 y5.339
0.743 y0.130
0.011 2–4 in family
UU UU
UU
Ž .
Ž .
Ž .
Female P average y0.025
0.003 y1.982
0.222 y0.016
0.002 number of children
UU UU
U
Ž .
Ž .
Ž .
Age 16 or 17 y0.223
0.007 y10.666
0.574 0.009
0.005
UU UU
Ž .
Ž .
Ž .
Age 18 or 19 y0.078
0.006 y6.182
0.441 0.003 0.004
UU UU
Ž .
Ž .
Ž .
Ages 20–24 y0.004 0.004
y2.192 0.284
0.028 0.002
UU UU
UU
Ž .
Ž .
Ž .
Ages 55–59 y0.201
0.005 y8.875
0.367 y0.220
0.003
UU UU
UU
Ž .
Ž .
Ž .
Ages 60–64 y0.272
0.021 y12.794
1.138 y0.269
0.017
UU UU
UU
Ž .
Ž .
Ž .
School qualification only 0.070
0.003 2.398
0.278 0.092
0.002
UU UU
UU
Ž .
Ž .
Ž .
Post-school 0.109
0.004 3.534
0.305 0.092
0.002 qualification only
UU UU
UU
Ž .
Ž .
Ž .
School and post- 0.159
0.003 6.417
0.291 0.139
0.002 school qualification
UU UU
UU
Ž .
Ž .
Ž .
University degree 0.168
0.005 6.993
0.382 0.169
0.003
UU UU
UU
Ž .
Ž .
Ž .
Rise in age of eligibility 0.051
0.014 2.334
0.662 0.038
0.012 for superannuation
UU UU
Ž .
Ž .
Ž .
Rise in age of eligibility y0.095
0.005 y5.695
0.215 0.006 0.004
for UB and DPB
UU UU
UU
Ž .
Ž .
Ž .
Log of real maximum y0.084
0.009 y3.151
0.454 y0.221
0.008 weekly benefit
UU UU
UU
Ž .
Ž .
Ž .
Log of real breakeven y0.200
0.015 y7.644
0.792 y0.199
0.013 income level
N 36,818
36,818 36,818
2
R 0.825
0.851 0.853
U
Significant at a 10 level, two-tailed test.
UU
Significant at a 1 level, two-tailed test.
earnings disregard or a rise in the benefit reduction rate increases labour force participation. A 10 cut in breakeven income raises participation by 2 percentage
points. As we said earlier, LFPRs may not adequately capture the overall labour supply
of the individuals in our sample, because they ignore any variation in the amount participants are willing to work. The regression model was re-estimated with a
new dependent variable on ‘‘weekly hours of labour supplied’’. These results are reported in the second column of Table 3. Note that the estimated coefficients and
their standard errors increase substantially in magnitude as we move between the first two columns in this table. The reason is that the ‘‘scale’’ of the dependent
variables has changed. Rather than talking about changes in the propensity to participate, we are now looking at changes in hours of labour supplied per week.
The estimated coefficients on the variables for increases in the ages of eligibility have the same signs and statistical significance levels as those found
earlier. The higher age of eligibility for Superannuation directly increased aggre- gate weekly labour supply, while the higher age of eligibility for UB and DPB had
the opposite impact.
The estimated coefficient on the log of real maximum weekly benefits is y3.151, and statistically significant at better than a 1 level. We can directly
compare this result to the earlier finding on participation by computing the relevant elasticity. Dividing this estimated coefficient by the mean of the depen-
Ž .
dent variable 28.421 , gives us an estimated elasticity of y0.111. The elasticities associated with both participation and hours of labour supplied are quite similar. A
10 cut in benefits increases weekly hours of labour supplied by 1.11. The most surprising results obtained thus far have been the estimated negative
effects of tighter eligibility criteria for UB and DPB on both participation and weekly hours of labour supplied. One reason for these unexpected findings may be
that these labour supply measures are excessively ‘‘narrow’’ for the affected group of 16- or 17-year-olds. On this basis, the regression was re-estimated with a
dependent variable that includes current participation in either the labour force or education. These results are reported in the third column of Table 3. The estimated
coefficient on the rise in the age of eligibility for UB and DPB is now positiÕe, but insignificant at a 10 level. This implies that labour force or educational
participation among 16- or 17-year-olds was directly unaffected by this tightening in eligibility criteria. Moreover, relative to the earlier results on the LFPR, the
estimated coefficient on the log of real maximum weekly benefits has increased
Ž .
substantially in absolute magnitude y0.221 , while the estimated coefficient on Ž
. the real breakeven income level is essentially unchanged y0.199 . The former
result suggests that deep cuts in potential benefits to those aged between 18 and 24 increased their participation in both the labour force and education.
These regression results provide empirical estimates of labour supply responses to both quantitative and qualitative changes in New Zealand’s social welfare
programmes. To put these responses in perspective, we multiply these estimated coefficients by the actual changes in these policy variables between 1990 and
1995. The resulting calculations are displayed in Table 4. Tighter eligibility criteria for UB, DPB and Superannuation had large labour
supply effects on the specific age groups directly affected by these policies, but relatively small effects on the overall labour market. For example, the rise in the
age of eligibility for Superannuation directly increased the aggregate LFPR between 1990 and 1995 by less than two-tenths of a percentage point. The rise in
the age of eligibility for UB and DPB lowered the aggregate LFPR by approxi- mately one-half of a percentage point, but had no measurable impact on overall
participation in either the labour force or education. These results suggest that the rise in the age of eligibility of UB and DPB from 16 to 18 has altered the
‘‘composition’’ of economic activity by reducing labour force participation in favour of human capital accumulation.
We estimate that the decline in real maximum weekly benefits between 1990 and 1995 resulted in increases of 0.82 percentage points in the aggregate LFPR,
0.3060 in average hours of labour supplied per week, and 2.15 percentage points in aggregate participation in the labour force or studying. The decline in real
breakeven income levels over the same period increased labour force participation by 1.07 percentage points, hours of labour supplied by 0.4074 per week, and
overall participation in the labour force or studying by 1.06 percentage points. The ‘‘net effects’’ of all of the welfare reforms between 1990 and 1995 on these
different measures of labour supply are shown in the bottom row of Table 4. We estimate that these overall benefit reforms lead to increases of 1.57 percentage
points in the LFPR, 0.5010 in average hours of labour supplied per week, and 3.32 percentage points in participation in the labour force or education. The vast
majority of these overall effects come from the changes in the benefit formula rather than changes in eligibility criteria.
Table 4 Ž
. Estimated effects of benefit reform on labour supply 1990–1995
The estimated policy effects in this table are the estimated coefficients in Table 3 multiplied by the actual changes in the mean policy variables between 1990 and 1995. The overall effects at the bottom
of each column are the summations of these individual policy effects.
Change in Change in
Change in participation
weekly hours participation
in the labour of labour
in the labour force
supplied force or education
Rise in age of eligibility for superannuation 0.0015
0.0687 0.0011
Rise in age of eligibility for UB and DPB y0.0047
y0.2811 0.0000
Decline in log of real maximum weekly benefits 0.0082
0.3060 0.0215
Decline in log of real breakeven income levels 0.0107
0.4074 0.0106
Overall effects of benefit reforms 0.0157
0.5010 0.0332
One ‘‘check’’ on these empirical results is to compare these estimated figures to the actual changes in these dependent variables between 1990 and 1995. In
other words, what proportion of the observed changes in these labour supply measures can be ‘‘explained’’ by the benefit reforms that took place over this
period? The simple answer is that nearly all of the observed increases in labour supply over this 5-year period can be attributed to these reforms. The actual
increases were 1.50 percentage points for the aggregate LFPR, 0.7195 for average hours of labour supplied per week, and 2.11 percentage points for aggregate
participation in the labour force or education.
Finally, to compare the labour supply responses across the three measures, we can convert these figures into elasticities by dividing by the means of the
respective dependent variables over the sample period. The overall estimated effects of these reforms were 2.10 for the LFPR, 1.76 for weekly hours of
labour supply and 4.20 for participation in either the labour force or education.
6. Conclusions
