Demand for outpatient healthcare

About 200 patients consulted more than one provider and a small number of individuals reported being not treated but consulting medicine shops and other non-medical practitioners. All of these observations were dropped from the analysis.

Illnesses associated with those seeking outpatient healthcare in rural areas are presented in table A1 of the Supplemental Digital Content.

Distance to the civic hospital and distance to the private hospital represented the distance variables for the public facility and the private facility, respectively.

Restricted activity in the survey referred to the state of health that prevented the ailing person from doing any of her/his normal activities, which varies by occupation (see NSSO[23] for the details).

Note that with the IV parameters unrestricted, the estimates are not invariant across scaling normalizations. This means that there is no obvious relationship between these two parameter estimates and the log-likelihood values at convergence are not identical (−17732.95 vs -17893.19). However, when the degenerate partition IV parameter is set to unity in RU2, this is consistent with the utility maximization hypothesis. But unrestricted random utility model 1 (RU1) or the non-normalized nested logit model estimate results are not consistent with the RU2 results. However, with the restriction that the IV parameters are equal in RU1, invariance is achieved across normalizations after accounting for scaling. The log-likelihood values at convergence are now equal (−17893.19) and the IV parameters are inverse to one another. Multiplying the utility function parameter estimates by the corresponding IV parameter estimates produces identical results.

The arc price elasticity was computed in the following way. First, the predicted probability of an individual choosing an alternative based on the preferred model specification was calculated. Second, the predicted probability of an individual choosing an alternative at a new consumption level due to an increased price was calculated using the simulation feature of the LIMDEP/NLOGIT 3.0 software package. Finally, the arc price elasticity was calculated as the percentage change in the sum of the average predicted probabilities divided by the average percentage change in price. The arc income elasticity was calculated using a similar methodology.

Note that while simulating the predicted probabilities for changes in price, some observations had to be dropped because net consumption turned out to be non-positive for some individuals. Increases in the price at public facilities of Indian rupees (Re)50, Re100, Re200 and Re300 resulted in the loss of 8, 22, 78 and 165 observations, respectively. The corresponding loss of observations due to the same increases in the price at private facilities and private doctors was 26, 70, 165 and 281 observations, and 24, 44, 118 and 224 observations, respectively.