I am having omission of regression coefficients due to collinearity in an event-study analysis. I am using district-level panel data.
For my Event study regression, I proceeded as follows:
1. I define two indicator variables, the first is for one year after treatment and the other is for one year before treatment. So, for a district in which treatment happened in 2006 say, I'm having an indicator for 2005 as being the year before treatment and 2007 as the year after treatment.
2. I am only using the three years of data for the district centered around its treatment year. So, sticking with the example of a district opening in 2006, I am using data from that district for the years 2005, 2006, and 2007 only. All other years for this district are dropped from the analysis. Hence, I have a balanced panel. Each district will have three years of data, but the three years included for each district will vary by district.
Here is the data sample:
Code:
* Example generated by -dataex-. To install: ssc install dataex clear input double district float year int treatmentyear byte(onebefore treatmentyearevent oneafter) double output 10311 2009 2008 0 0 1 10997127 10311 2008 2008 0 1 0 10263347 10311 2007 2008 1 0 0 7157395 10313 2009 2008 0 0 1 32174431 10313 2008 2008 0 1 0 31922877 10313 2007 2008 1 0 0 20902644 10511 2008 2007 0 0 1 5190128 10511 2007 2007 0 1 0 2483258 10511 2006 2007 1 0 0 1543567 10515 2008 2007 0 0 1 2467145 10515 2007 2007 0 1 0 1218880 10515 2006 2007 1 0 0 622912 10517 2008 2007 0 0 1 9245827 10517 2007 2007 0 1 0 5592740 10517 2006 2007 1 0 0 3781691 10521 2009 2008 0 0 1 6452216 10521 2008 2008 0 1 0 4272063 10521 2007 2008 1 0 0 2266553 10527 2012 2011 0 0 1 2285181.7995331287 10527 2011 2011 0 1 0 2159072.699913025 10527 2010 2011 1 0 0 1754932 10701 2004 2003 0 0 1 4073798 10701 2003 2003 0 1 0 3215059 10701 2002 2003 1 0 0 2270777 10703 2007 2006 0 0 1 1448312 10703 2006 2006 0 1 0 649834 10703 2005 2006 1 0 0 595683 10711 2007 2006 0 0 1 570822 10711 2006 2006 0 1 0 501945 10711 2005 2006 1 0 0 274273 10717 2006 2005 0 0 1 2415248 10717 2005 2005 0 1 0 2620946 10717 2004 2005 1 0 0 2125851 10903 2008 2007 0 0 1 4548873 10903 2007 2007 0 1 0 3447862 10903 2006 2007 1 0 0 2313126 10905 2009 2008 0 0 1 771627 10905 2008 2008 0 1 0 439136 10905 2007 2008 1 0 0 436146 10909 2011 2010 0 0 1 1105832.5000915527 10909 2010 2010 0 1 0 319740 10909 2009 2010 1 0 0 528248 11101 2007 2006 0 0 1 3771011 11101 2006 2006 0 1 0 2764468 11101 2005 2006 1 0 0 2171111 11103 2008 2007 0 0 1 2639729 11103 2007 2007 0 1 0 1506677 11103 2006 2007 1 0 0 733102 11301 2003 2002 0 0 1 4940093 11301 2002 2002 0 1 0 4377711 11301 2001 2002 1 0 0 3666350 11303 2009 2008 0 0 1 427574 11303 2008 2008 0 1 0 370241 11303 2007 2008 1 0 0 229208 11317 2016 2015 0 0 1 2179215 11317 2015 2015 0 1 0 2349366 11317 2014 2015 1 0 0 2145403.699951172 11501 2003 2002 0 0 1 2841519 11501 2002 2002 0 1 0 2221096 11501 2001 2002 1 0 0 1902339 11503 2011 2010 0 0 1 6064693.299743652 11503 2010 2010 0 1 0 4796533 11503 2009 2010 1 0 0 1403027 11509 2007 2006 0 0 1 845161 11509 2006 2006 0 1 0 506974 11509 2005 2006 1 0 0 404228 11701 2011 2010 0 0 1 26754995 11701 2010 2010 0 1 0 21081781 11701 2009 2010 1 0 0 12932518 11703 2009 2008 0 0 1 7338138 11703 2008 2008 0 1 0 12183892 11703 2007 2008 1 0 0 5549341 11707 2010 2009 0 0 1 6512476 11707 2009 2009 0 1 0 1487311 11707 2008 2009 1 0 0 869042 11713 2012 2011 0 0 1 3211529 11713 2011 2011 0 1 0 1980381 11713 2010 2011 1 0 0 1110997 20103 2017 2016 0 0 1 72258 20103 2016 2016 0 1 0 100539.60000610352 20103 2015 2016 1 0 0 58667 20115 2007 2006 0 0 1 240739 20115 2006 2006 0 1 0 108612 20115 2005 2006 1 0 0 84235 20301 2016 2015 0 0 1 7219632.999850273 20301 2015 2015 0 1 0 6591344.000009179 20301 2014 2015 1 0 0 7412385.699951172 20501 2003 2002 0 0 1 1953630 20501 2002 2002 0 1 0 1520629 20501 2001 2002 1 0 0 1125519 20511 2012 2011 0 0 1 4908541.599914551 20511 2011 2011 0 1 0 3874276.299987793 20511 2010 2011 1 0 0 1877709 20917 2015 2014 0 0 1 1177131.4000015259 20917 2014 2014 0 1 0 1064612 20917 2013 2014 1 0 0 691924 21101 2007 2006 0 0 1 1768123 21101 2006 2006 0 1 0 1399262 21101 2005 2006 1 0 0 1189413 21301 2009 2008 0 0 1 4403553 end
Using that sub-sample of data, I am running the regression
Code:
reghdfe Y onebefore oneafter, absorb (district year) cluster(district)
When I run this regression, the coefficient on oneafter is being omitted. Is it logical? I did not expect it as I am already omitting the year of treatment indicator and I do not have any district where all the three event indicators have same value. Is there a way to fix this?
When I run with 5 year-year period centered around the zone opening year, one year is still being dropped despite the fact that I am not including the treatment year indicator.
Thank you,
Marina
0 Response to Collinearity in event study analysis
Post a Comment