5.2  Marginal effects

It is difficult to derive the effect of characteristics on labour supply directly from the estimated model. The results of the model can be summarized by calculating the expected hours of labour supply and the probability of non-participation. To facilitate the interpretation of the results, Table 4 presents the average expected level of labour supply and non-participation for the five subgroups when one of the characteristics is held fixed at a specific value, while all other characteristics are as observed in the sample. This allows us to isolate the average effect of a change in one particular characteristic across the samples.

Note a: Expected labour supply is calculated at the sample characteristics except for the variable named in the first column of the row, which is set to one for all individuals and the dummy variables for the other categories of the same variable are set to zero. Except for the age variable, which is a continuous variable and increased by 10 per cent for all individuals in the sample. Average unweighted expected hours of work are computed across the samples.

b: This subgroup is quite small for sole parents at just 18 individuals.

The first rows in the two sections of the table give the average expected labour supply and non-participation in the labour force across the unweighted sample for all demographic groups. The predicted values are similar to the observed values. The second rows in the two sections, presents the effect of an increase in wage levels by 10 per cent for all individuals. For all groups, the expected hours worked increases and expected labour force participation increases. The average wage elasticities implied by the effect are 0.63, 0.82 and 0.34 for single men, single women and sole parents respectively. For married men and women, the wage increase of the partner also affects their labour supply. The elasticity of changing both wage rates by the same percentage is 0.24 and 0.40 respectively. The own wage elasticity is expected to be somewhat higher, because when only their own wages are increased there would not be a counteracting income effect from their partners’ wages.

The next two rows look at the effect of having a child aged between 1 and 3 years old compared to not having a child aged less than 9 years. This is of course only relevant for couples and sole parents. The table shows clearly that married men are much less affected by the presence of children than married women or sole parents, although a slight reduction in labour supply is visible. Education also has a much larger effect for women and sole parents than for men, but the effect of education on male labour supply is higher than the effect of having children for men. Similarly, the effect of partner’s education is more pronounced for married women than men. An individual’s own age has a relatively larger effect on female labour supply than on male labour supply and is higher for couples than for single-adult households. Partner’s age appears much less relevant. Finally, amongst sole parents the effect of being male is to work more hours and be slightly more likely to participate. The effects found here are quite similar to the results found by Maloney (2000) who used more highly aggregated data, fewer variables, a reduced form model, and did not estimate separate models for subgroups.[22]

To explore the effect of the change in the eligibility for the New Zealand Superannuation, the sample is restricted to those who are 60 or older. The predicted labour supply is shown in the relevant row and it is obvious that overall labour supply is much lower for this group. For this subgroup, the indicator for being eligible for the Superannuation is put to zero and one respectively. This clearly shows the relatively large effect of being eligible, which is at least as high for men as for women, if not higher. The effect of the partner’s eligibility is larger for men as well. As noted before, the effect is a mixture of income and direct preference effects. The available income at the different hours points was not changed in this calculation, which means the numbers in Table 4 are likely to underestimate the effect.