logit treat mtbv lgta_cpi cflowr1 tdebtr capxr
set seed 123456
. gen u=uniform()
. sort u
.
. *
. psmatch2 treat, pscore(pscore) outcome(roa) common radius caliper(0.01)
There are observations with identical propensity score values.
The sort order of the data could affect your results.
Make sure that the sort order is random before calling psmatch2.
Variable Sample Treated Controls Difference S.E. T-stat
roa Unmatched .081625289 .111897945 -.030272656 .017641929 -1.72
ATT .081625289 .095503262 -.013877973 .032015272 -0.43
Note: S.E. does not take into account that the propensity score is estimated.
question 1: I interpret above results as the average treatment effect is -.013877973. Do I need to use T-stat to determine if the estimation is statistcially significant?
question 2: prior study I try to replicate "select matched firms (with replacement) with the propsensity score within +/-1% of the each other from the treated firms. am i using the right specification ?'
psmatch2 treat, pscore(pscore) outcome(roa) common radius caliper(0.01)
the balancing test, the documents I google search used
pstest mtbv lgta_cpi cflowr1 tdebtr capxr , sum
Mean t-test V(T)/
Variable Treated Control %bias t p>t V(C)
mtbv 1.4549 1.5017 -3.8 -0.20 0.844 1.03
lgta_cpi 4.297 4.5191 -11.9 -0.67 0.501 0.62
cflowr1 .07558 .09635 -10.8 -0.61 0.545 1.52
tdebtr .22123 .2101 6.2 0.39 0.695 0.92
capxr .06089 .05581 7.2 0.62 0.534 1.63*
* if variance ratio outside [0.62; 1.60]
Ps R2 LR chi2 p>chi2 MeanBias MedBias B R %Var
0.008 1.57 0.904 8.0 7.2 20.9 0.77 20
* if B>25%, R outside [0.5; 2]
question 3: should I be using pstest for balancing test?
question 4: does the results above suggest that the bias are reasonable ? I thought there is <5% rule .
thanks,
Rochelle
