model代写 When including both the variables in the model, privateins is proved to be insignificant, but publicins is proved to be significant.
Taking logarithm of h_expend as the dependent variable, and all other variables available except profession and sectoral variables and flu_vac and use_seatb as the explanatory variables, we did regression on this model. As an result, variables female and pregnant are omitted because of collinearity; under 5% significance level, coefficients of some variables are not significantly different from 0, coefficients of variables duid, pid, insur_fr, age, civ_stat, hisp, n_comb, ph_perc, mh_perc, smoker, fam_size, region, p_weight, and kessler are significantly different from 0. model代写
Apprently, duid, pid, insur_fr, age, civ_stat, hisp, n_comb, ph_perc, mh_perc, smoker, fam_size, region, p_weight, and kessler might be the potential exogenous explanatory variables, and theoretically the variable insur should be in the model, because it may affect the patients’ and doctors’ joint decisions about treatment plans, thus affecting healthcare expenditure. However, if we run the regression on variables insur, insur_fr, n_comorb, ph_perc, mh_perc , smoker, p_weight, Kessler, we see coefficients of them are all significantly different from zero, except for insur.model代写
The variable insur represents three status of being insured (1: privately insured, 2: publicly insured, 3: not insured), which will cause problems because for private insurance, the healthcare expenditure will depend on the information that the insurance taker has but not the insurance company. In order to avoid information asymmetry, the private-insurance shouldn’t be included in the model.model代写
Using the commands gen privateins=(insur==1) and gen publicins=(insur==2) to generate two dummy variables and represent private insurance and public insurance respectively.
When including both the variables in the model, privateins is proved to be insignificant, but publicins is proved to be significant. If we only include publicins in the model, the variable civ_stat also becomes insignificant.model代写
As insur might be influenced by other explanatory variables in the model when it is equal to 2, it might be endogenous.
An instrumental variable must (1). be correlated with the variable that it replaces, (2). not correlated with the error term, (3). not correlated with other exogenous variables. (4). And if more instrumental variables are needed, they should not be correlated with each other.model代写
As mentioned by the text, profession and sectoral variables and flu_vac and use_seatb might be instruments. Obviously, these variables are not correlated with each other. If we do regression for residuals on these variables, we see that except OCCCAT31, all other potential instruments are not correlated with the residuals. And variables EMPST42, EMPST53, MORJOB31, MORJOB42, MORJOB53, SELFCM31, INDCAT31, OCCCAT42, OCCCAT53, flu_vac, use_seatb are all related with the variable insur, but not related with other variables. Thus, some of them meet the above requirements.model代写
Firstly, doing regression for insur on all the possible instruments EMPST31, EMPST42, EMPST53, MORJOB53, ELFCM31, flu_vac, use_seatb, we see some instruments are significant. Then we get the residual z_res, and do the regression for loghe on publicins, exogenous variables and z_res, we see the coefficient of z_res is significantly different from 0. Thus OLS is not consistent.
As privateins is proved to be insignicant in this model, here we don’t need to check the endogeneity of this variable.model代写
reg loghe EMPST42 SELFCM31 INDCAT31 OCCCAT42 OCCCAT53 insur_fr age hisp n_comorb ph_perc mh_perc fam_size region p_weight kessler
We see the coefficient of EMPST42 is significantly different from 0, thus it has direct influence on the healthcare expenditure.model代写
ivregress 2sls loghe insur_fr age hisp n_comorb smoker p_weight kessler (publicins=EMPST31 EMPST42 EMPST53 SELFCM31 use_seatb)
Comparing the two results shown above, the estimates and standard errors are not very different.model代写
Doing the first stage regression, we do the F-test for each instrument. We can conclude that only INDCAT31 is weak, all other instruments (EMPST42, SELFCM31, OCCCAT42, OCCCAT53) are strong.model代写
Please check the answer of question 4a. Since the coefficient of z_res is significantly different from 0, insur is endonegous.
We generate 3 variables dSE31, dIN31 and dOC42 which are equal to 2 multiplying SELFCM31, INDCAT31 and OCCCAT42 respectively. We add them in the first stage regression, then perform F-test, we see F-statistic of all these variables are larger than 10, which means they are all strong.model代写
If we use n instruments, they will be strong but have no effect on the regression.
This means the variable publicins is exogenous since it has no relationship with any of explanatory variables in the model.
If we treat this group differently, we should replace the variable insur with publicins and privateins, and select proper instruments for privateins.model代写
Because the model will stop running when it converges.
The marginal effect of the Tobit model is bigger that of my previous models, which means 1% change in the explanatory variables will cause bigger change in the dependent variable for Tobit model.model代写