Research design question

[Maria Chr]

Hello,

I am researching the effect of an increase in the maternity leave duration on wages of mothers using panel data. I initially thought of using mothers of infants ( below the age of 1) as my treatment group as they are the women who are directly affected by the extension and as my control group mothers of children over the age of 5 which are unlikely to become pregnant.

However I have noticed that due to the panel structure of my data which runs from 1991 to 2009, the infants at some point grow and become 5 years old. Therefore there are some women that are both in my treatment and control group. This makes me think that my research design is problematic. Could somebody please advise on what I could do to correct this?

[Phil Bromiley]

While it sometimes happens that folks get such research design advice on this list, it really is more directed at implementing design than doing design.

You're not even clear about what estimator or analysis you intend to run. You talk about control groups but don't tell us are you going to run a treatment model, panel data model or what.

The answer to your question really depends on the detailed understanding of the context. Do you want to assume that the increase in maternity leave only influences the compensation for those who are actually receiving it, or could it influence compensation for all women might receive it, or even for all women in general?

[Maria Chr]

Hello Phil and thank you very much for your response.

Apologies for not being clear.

I would like to assume that the increase in maternity leave influences the compensation for all women who might receive it, i.e. all eligible women. Eligibility for my reform means that women have been employed while pregnant. I have managed to restrict my sample to include all of these women.

I am trying to run a difference-in-difference regression that will allow me to compare the wages of women who have given birth before the reform (i.e. before 2000) to the wages of women who gave birth after the reform (i.e. after 2000).

However, given that I am using panel data from 1991 to 2009 which interviews the same women from one year to another I am really not sure how to correctly define my treatment and control groups.

Any suggestions, please?

[Carlo Lazzaro]

Maria:
can't you focus on women's first birth and compare the situations before and after the reform?

[Maria Chr]

Carlo:

Thank you for your response.

In that case, what would be my control and treatment group in a diff-in-diff methodology?

[Carlo Lazzaro]

Maria:
does you sample include more than one country?

[Maria Chr]

Carlo:

No it's just a single country : the UK and it's panel data whereby individuals are surveyed annually. It also contains the exact date of birth of the children.

[Carlo Lazzaro]

Maria:
the idea might be to perform a panel data regression (an not a difference in difference regression) that consider women's wage as regressand and, among predictors, a before/after reform categorical variable that splits the dataset in before/after reform eras. I would restrict the sample to women's first birth.

[Maria Chr]

Carlo:

I will try to do that. Thanks a lot for your help!