I am currently writing a seminar paper. I am estimating the effect of the share of leftwing government members in swiss cantons on the public expenses (and their categories) of these cantons.
The data is balanced panel data with N (cantons) = 26 and T(years) = 28.
The dependent variable is: public expenses in category j per canton in year t
The main independent variable is: share of left-wing government members in government of canton i in year t.
(Category means for example: health, social security, culture, education, etc.)
As control variables I have:
- GDP per canton
- unemployment rate per canton
- ratio of people aged > 64 per canton
- debt per canton
- population per canton
- and a lagged variable of the proportion of left-wing politicians in the parliament
Here is a sample of my main data for one canton:
Code:
* Example generated by -dataex-. To install: ssc install dataex clear input int jahr long kanton double(links total BIP) long bev double(alt schulden Alquote) 1990 1 .2 2103595.49679 22258.61295548331 498035 .126 4118853.07339 .0021 1991 1 .2 2418372.7013 23247.730242745427 506818 .126 4258530.37578 .0055000000000000005 1992 1 .2 2622013.84704 23731.27671978167 511979 .126 3995904.96175 .01648507796021844 1993 1 .2 2752637.76148 24251.730934888055 518945 .126 4020907.67093 .03353853343994975 1994 1 .2 2866913.15304 24850.55070690186 523114 .127 4312582.93007 .032740127331169364 1995 1 .2 2871594.22682 25154.26712852215 528887 .12800000000000003 4261014.9935 .02906240833847551 1996 1 0 3033372.5041 25331.135728586345 531665 .129 4566428.24205 .03839377973484641 1997 1 0 3065249.40363 25817.39558783384 534028 .13 4734656.4063 .046613346283090946 1998 1 0 3092004.07619 26558.17208651548 536462 .132 4797195.89518 .030217723885365436 1999 1 0 3184180.88853 27039.352615536056 540639 .13195681406631782 4907276.22711 .021088996722396877 2000 1 0 3288728.78046 28525.767132413897 544306 .13364541269065564 4901166.57689 .013503121668950816 2001 1 0 3448772.61443 29194.49554783396 550298 .13510134508938793 4762340.0067 .01214187822228023 2002 1 0 3559261.63535 29167.80471578718 555782 .1365679349097308 4489301.77899 .02123115577889447 2003 1 0 3771771 29508.175608019385 560674 .13796074010922568 4366114.90529 .033257738911005245 2004 1 0 3868402.48256 30431.545233146422 565122 .14008302631998046 5305308.83576 .034339989993256326 2005 1 0 3987495.90309 31596.1406982345 569344 .14222684352517986 4762380.78077 .03251647849637799 2006 1 0 4304953.29254 33544.994154576845 574813 .14474272502535607 4510972.05203 .02857251626095847 2007 1 0 4505096.31065 35767.62753463508 581562 .147655795942651 4886900.96703 .023552013313319846 2008 1 .25 3995789.07397 37774.52858 591632 .15066122184060363 4832603.85848 .022927679523156913 2009 1 .4 4107370.48387 36943.54089 600040 .1535130991267249 4584035.66422 .033850257782418576 2010 1 .4 4146058.66746 37664.96637 608299 .15526129007990633 4696771.97718 .03128852310932995 2011 1 .4 4336101.42091 38505.31274 618298 .15872281650595668 4311897.59296 .025658837672748246 2012 1 .4 4577403.78175 38719.76854 627340 .16137022985940638 3900244.74448 .02685290486325758 2013 1 .4 4703960.67493 39488.54672 636362 .1640371360954928 3744099.61901 .028494090775842893 2014 1 .4 4770144.18455 40139.27705 645277 .1663657623005314 4051689.93526 .027857224266495634 2015 1 .4 4892710.96136 40647.58003 653675 .16873522775079358 4254435.63835 .029879852698914817 2016 1 .2 4973509.49627 40813.49 663462 .1705282291977536 4838286.30867 .031555575596547404 2017 1 .2 4944214.42435 41592.47817 670988 .17351129975498816 4990070.88585 .030368726678589565 end label values kanton kanton1 label def kanton1 1 "AG", modify label var jahr "jahr" label var kanton "kanton" label var links "Anteil links in Regierung" label var total "Total Ausgaben" label var BIP "BIP" label var bev "Total Wohnbevölkerung" label var alt "Anteil Bevölkerung >64" label var schulden "Bruttoschulden" label var Alquote "Arbeitslosenquote in Dezimalzahlen"
My question to you is now: Should I use the growth rates in the variables (the dependent variables as well as the control variables) or their levels? I decided to estimate a LSDVC model (xtlsdvc in STATA). When I use levels, I see some effects, but when I use growth rates, virtually all the variables become insignificant.
I use the following code in STATA for the analysis in levels:
Code:
xtlsdvc kat03 links lagd_mlp bev alt schulden Alquote BIP, initial(ab) vcov(50) first
Code:
xtlsdvc gln_kat03 links lagd_mlp gln_alt gln_schulden gln_Alquote gln_BIP, initial(ab) vcov(50) first
The data are stationary and the Hausman-test result proposed to use fixed effects.
Also: Do you think I am estimating the right model?
Thank you so much, your answer would help me an awful lot!!
Regards,
Lara Knuchel
0 Response to Panel Data Analysis: Growth rates or levels?
Post a Comment