The Example of Employer-Supported Childcare
The author asks how far the extension of employer-supported childcare serves as a driver for higher maternal labor supply. She addresses this question by categorizing employer-supported childcare as an efficiency wage introduced by the employer to increase the working volume of mothers. Applying various impact evaluation techniques in an econometric analysis, the author concludes that the availability of employer-supported childcare has a positive impact on the length and working volume of mothers who return back to work after giving birth. Furthermore, the usage of employer-supported childcare by mothers with pre-school age children influences the amount of agreed and actual working hours positively.
Panel data includes subsequent observations for individuals, which allow the analysis of changes over time. Thereby, one is able to control for variables, which change across observations, but not over time (for instance gender). Furthermore, the data allows the consideration of factors which cannot be included for practical reasons. For instance, unobservable factors like cultural attitudes differ from one individual to another individual, but less likely over time. Consequentially, changes in the dependent variables cannot result in omitted variable bias, under the condition that the variable does not change over time. The technique for controlling this challenge within regression frameworks is called FE.
Basically, using FE models addresses the assumption implying that features of the observations could impact the dependent variable, which in turn might bias the results by removing the effect of those characteristics from the dependent variable. The starting regression framework for FE is
Yit = β0 + β1 Xit + β2 Zi + εit (1)
Whereby Yit denotes the dependent variable, Xit denotes the independent variable and εit is the error term, including the subscript i refers to number of individuals and t refers to the time periods. Zi shall exemplary demonstrate an unobservable variable changing across individuals, but not over the time periods. The goal is to acquire an estimate for β1, which can be interpreted as the effect of X on Y, holding constant Z. This intended interpretation can also be rewritten as having individual intercept n for each state
Yit = β1 Xit + α + εit (2)
whereby here αi = β0 + β2 Zi going from 1 to ni represents the intercept for each individual observation. As a result, the slope coefficient is the same for all individuals, however it is different and it is not correlated with the intercepts between the individuals (Stock & Watson, 2012, p. 397).
Using FE, the following assumptions remain. Firstly, the conditional mean of the error term in the present and past is zero. As already indicated above, this implies that there are no omitted variables biases. Secondly, it is assumed that simple random sampling is applied to randomly select the sample from the population. It is however not required that the entities are identically and independently selected over time. The third assumption implies that Xit and Yit have finite fourth moments. Fourthly, it is assumed there is no perfect multicollinearity. ← 213 | 214 →