control variables whose effects are allowed to vary by year

[Nicola Meier]

Dear Statalist, I'm quite new to stata and want to investigate the effect population ageing has on economic growth in Japan similar to the work of Maestas et al. (2016).
Therefore, I calculate the growth rates for GDP per capita and the fraction of people aged 65+ from 1995 to 2000, from 2000 to 2005, from 2005 to 2010 and from 2010 to 2015. I do this for all 47 Japanese prefectures and after that I want to run an OLS and an IV regression. Now the problem: I want to include the log of the fraction of persons working in the three major industries as a control variable and the control variable's effect should be allowed to vary by year. My question is: how can I allow the control variable in Stata to vary by year?

[Clyde Schechter]

So you want to include interaction terms between your time period variable and these industry fractions. Let's say your time period variable is called era and takes on the values 1 for 1995-1999, 2 for 2000-2004, 3 for 2005-2009, and 4 from 2010-2014 (the groupings you show in your description make no sense as years 2000, 2005, and 2010 would belong to more than one time period), and let's call the log industry fraction variables ind_frac1 ind_frac2 and ind_frac3. Then you would include in your model

i.era##c.(ind_frac1 ind_frac2 ind_frac3)

Read -help fvvarlist- for more information about this factor-variable notation. Also, you will probably want to use the -margins- command to interpret the results you get from the regressions. I think the best introduction to the -margins- command is the excellent Richard Williams' https://www3.nd.edu/~rwilliam/stats/Margins01.pdf.

[Nicola Meier]

Thanks a lot! Your answer helped me a lot!