I have insurance claims data (health insurance) from which I'm trying to figure out what effects a remote / digital service offered to insurance customers has on the claims amount. This is a bit tricky is because the service is voluntary so selection bias is quite big problem and I don't have too many variables in data which would explain the usage of the service. The health insurance is also voluntary, we have large public sector, but some people opt to get private health insurance.
dependent variable:
- logCLAIMS (euros)
independent variables are:
- age (continuous)
- three class factor variable: where the treatment was given: 1. preferred provider organization 2. not PPO 3. public hospital
- gender(MALE, FEMALE, UNBORN CHILD > NO GENDER)
- treated variable is a dummy 0 / 1 in which 0 = did not use the service and 1 = did use the service
Instruments:
- area of living (certain areas of the country there are bigger cities and some mainly country side -> significant difference to the use of service)
- three class factor variable, where the treatment was given
My instruments are quite weak but this is all i can get atm from my data.
Questions:
- I would like to get your comments on the model and is there a better way to do this?
- I get ATE -0,75 (from eregress and etreg -models) and I'm wondering how margins can give me a result of 0,17, what can I interpret from here? ?
codes below
Code:
eregress logCLAIMS AGE i.b2.TREATMENTPLACE i.b2.GENDER, entreat (REMOTE = i.b2.TREATMENTPLACE i.b2.AREA, nointeract) vce(robust) Extended linear regression Number of obs = 17,084 Wald chi2(6) = 2274.00 Log likelihood = -25883.81 Prob > chi2 = 0.0000 ------------------------------------------------------------------------------------------ | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------------------+---------------------------------------------------------------- logCLAIMS| AGE| .0048115 .0002871 16.76 0.000 .0042488 .0053741 | TREATMENTPLACE| PPO| .264882 .0170253 15.56 0.000 .2315131 .2982509 NON PPO | 0 (base) PUBLIC HOSPITAL | -.7805487 .0243844 -32.01 0.000 -.8283414 -.7327561 | GENDER| NG | .0343502 .0463053 0.74 0.458 -.0564064 .1251069 M | 0 (base) F | -.0331972 .0128694 -2.58 0.010 -.0584206 -.0079737 | REMOTE | 0 | 0 (base) 1 | -.7573412 .05366 -14.11 0.000 -.8625128 -.6521696 | _cons | 4.873668 .0166892 292.03 0.000 4.840957 4.906378 -------------------------+---------------------------------------------------------------- REMOTE | TREATMENTPLACE| PPO| 1.314718 .0519358 25.31 0.000 1.212926 1.41651 NON PPO| 0 (base) PUBLIC HOSPITAL| -.203226 .1063506 -1.91 0.056 -.4116694 .0052174 | AREAS | 1 | -.0764202 .2209754 -0.35 0.729 -.5095241 .3566837 2 | 0 (base) 3 | -.0710966 .0456614 -1.56 0.119 -.1605913 .0183981 4| -.1733018 .0992317 -1.75 0.081 -.3677923 .0211887 5 | -.0832831 .0390727 -2.13 0.033 -.1598641 -.0067021 6 | .0906437 .0350018 2.59 0.010 .0220414 .159246 7 | .1551398 .0477874 3.25 0.001 .0614782 .2488014 | _cons | -2.261838 .0511611 -44.21 0.000 -2.362112 -2.161564 -------------------------+---------------------------------------------------------------- var(e.logCLAIMS)| .726331 .0109492 .7051849 .7481112 -------------------------+---------------------------------------------------------------- corr(e.REMOTE,| e.logCLAIMS)| .5082226 .0292915 17.35 0.000 .4485854 .563354 ------------------------------------------------------------------------------------------
but
margins, dydx(REMOTE)
Code:
Average marginal effects Number of obs = 17,084 Model VCE : Robust Expression : mean of logCLAIMS, predict() dy/dx w.r.t. : 1.REMOTE ------------------------------------------------------------------------------ | Delta-method | dy/dx Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- REMOTE 0 | 0 (base) 1 | .1748854 .0660175 2.65 0.008 .0454935 .3042772 ------------------------------------------------------------------------------ Note: dy/dx for factor levels is the discrete change from the base level.
0 Response to eregress / etregress and margins + interpretation
Post a Comment