I'm working on a dynamic Difference-in-Differences (DID) analysis to assess the impact of child care leave on individuals' wages over their working life. My database includes information on the date of joining and leaving the company, the date of leaving, and the cause for leaving (including child care leave). I also have data on the date of birth of cohabitants, helping identify when a child is born in the household.
Here's a concise overview of the scenario:
Leave can be taken anytime before the child turns 3, with no minimum duration. The maximum leave duration is until the child turns 3. I plan to use individuals who had a child but did not take home care leave as the control group, comparing them with those who took the leave (treatment group). However, I'm concerned about the assumption of parallel trends. For instance, individuals taking leave might have partners with higher wages (I cannot observe them), influencing their decision to take leave. This might lead to differences in annual income trends over time due to less incentives of the treatment group to work when having a child(if we suppose than on average they have partners with higher income). Considering I only have annual income data and cannot calculate hourly wages, I'm exploring the possibility of including anual days worked and type of contract (part-time or full-time) as controls.
Questions:
Would incorporating controls like days worked per year help in assuming parallel trends, considering that these variables may be influenced by the leave policy?
Given the staggered adoption of leave in the first, second, or third year since childbirth, I'm contemplating using an estimator robust to staggered adoption heterogeneity (e.g., Callaway and Santanna). However, the parallel trends issue persists. Any insights on addressing this concern?
I appreciate any guidance on refining my approach or alternative strategies to ensure a robust analysis.
My question is theoretical, so I haven't attempted to implement anything in any program yet. My intention with this question is to better understand if including controls such as days worked per year is a mistake in this context.
Comment