I could really use some advice on how to handle the data I have for a piece of research I am starting to work on. I am interested in trying to determine whether the Black Lives Matter protests (in particular violent ones) of last year had an impact on the November US presidential election.
Here is a small sample of the dataset I have:
Code:
* Example generated by -dataex-. To install: ssc install dataex clear input int EVENT_DATE str20 state str27 county str37 LOCATION long(rainfall county_fips) float(viol_prot diff_dem_2016_2020) . "Alabama" "Autauga" "" . 1001 0 3.061512 22205 "Alabama" "Baldwin" "Orange Beach" 0 1003 0 2.8437195 22101 "Alabama" "Baldwin" "Orange Beach" . 1003 1 2.8437195 22165 "Alabama" "Baldwin" "Orange Beach" . 1003 0 2.8437195 22149 "Alabama" "Baldwin" "Foley" 0 1003 0 2.8437195 22080 "Alabama" "Baldwin" "Fairhope" . 1003 0 2.8437195 . "Alabama" "Barbour" "" . 1005 0 -.8720779 22079 "Alabama" "Bibb" "Centreville" . 1007 0 -.7237587 . "Alabama" "Blount" "" . 1009 0 1.0994759 . "Alabama" "Bullock" "" . 1011 0 -.3884811 end format %td EVENT_DATE
As you can see, there are repeated observations for some of the counties (i.e. more than one protest occurred in that county between May and November). But of course, the dependent variable, relative to the election results, is the same for each observation by county. So my question is whether this could be a problem for estimation purposes, meaning the fact that I have a constant dependent variable relative to multiple observations (protests happened more than once but the election took place only once, in November). Is the data structure ok?
Moreover, when trying to run simple OLS regressions (my intention is to use more sophisticated methods down the road), I notice that when including county fixed effects and time fixed effects (relatively to the date of the protest)
Code:
reg diff_dem_2016_2020 viol_prot i.EVENT_DATE i.county_fips
Any help would be greatly appreciated.
0 Response to Doubt on data structure/fixed effects
Post a Comment