Ordinal logistic regression analysis of capasity use at health institutions poviding in-bed treatment

Estimation of logistic regression, which is one of the qualitative response models giving maximum possibilities to categories of response variable, requires certain conjectures. Ragarding these conjectures, in cases where the response variable is categorically ordinal and if the estimated ordinal logistic regression model is statistically significant, this suggested model can be used in responsive measures. It this study, we have used data from 2005 statistical manual of Turkish Republic Minestry of Health Institutions providing in-bed treatment. From these data, bed capacity rates of the mentioned institutions have been categorized as low, middle, and high and response of low capacity hospitals to middle and high capacity hospitals have been examined. Similarly, hospitals have been grouped according to bed numbers as small, medium, and large and responses of service given at small hospitals to services given at medium and large hospitals have been estimated. Institutional status of hospitals have also taken into consideration and responses of services given at private hospitals to services given at public and education-research hospitals have been compared. For the mentioned category estimation function, on the basis of internal descriptive variable categories ordinal logistic regression model was set where numbers of bed, practitioners, specialized doctors, polyclinics, hospitalized patients, days spent at hospital have been used as continuous descriptive variables; on the other hand capacity usage rate, size of hospital compared to number of beds, and finally institutional status of hospital have been taken as discontinuous variables. Considering category logits of middle and high levels of this conceptual model, results of the model have been estimated and tested. By comparing the models, possibilities of services given at low level hospitals and upper level hospitals have been shown by logit values of the mentioned models.