We examined the effect of salary on residents’ hospital choices. The results of pooled OLS were similar to those of Kawamura’s previous results in that the coefficient of salary was positive and not significant . However, the results of the fixed effect model were statistically significant, which were better supported by statistical specification tests. Basic characteristics of hospitals, such as hospital function, location, and ownership, are supposed to determine the attractiveness of hospitals and the level of income simultaneously [30, 43]. Therefore, results that control for this unobserved heterogeneity and endogenous relationship between hospital characteristics and salary level should more precisely identify the effect of salary on residents’ choice of training hospitals.
Previous studies have also indicated that financial incentives would affect physicians’ choice of training hospitals or workplace in high-income countries as well as in middle-low income countries [13–15, 27, 28]. A systematic review on human resource allocation in healthcare reported that increased wage for trainees may be an effective strategy to improve supply as well as distribution of physicians, though the validity of evidence was limited due to poor study design . With a sophisticated statistical technique and use of panel data, our results have added a strong support to the statement that monetary incentive is a significant factor for a physician’s choice of early career, especially in rural areas.
Our results also showed that the number of teaching physicians and that of inpatients had a positive effect on hospital choices. Previous studies also indicated that the quality of training environments, such as teaching skills of attending physicians and opportunities to learn clinical skills, was a significant factor that affected resident’s choice of specialty and training location, and job satisfaction [25, 27, 44, 45]. Although the number of teaching physicians does not necessarily guarantee the quality of teaching, busy attending physicians were reported to decrease residents’ satisfaction with the quality of attending teaching in the US national survey of surgical residents . The questionnaire survey by the MHLW in Japan also reported that in their choice of training, non-university residents took into consideration hospital factors, such as the number of clinical cases, to provide a sufficient learning experience (43.3%), the comprehensiveness of the training curriculum (37.5%), the number and quality of teaching physicians (29.7%), and high-tech therapeutic equipment of hospital facilities (27.3%) . These conditions are expected to provide young residents with better opportunities to enhance their clinical skills and wider alternatives for career development.
In the current study in the analysis stratified by city size, the interaction term was statistically significant and negative in both groups, which suggested that the effects of salary and the number of teaching physicians were antagonistic. Therefore, the effect of salary was diminished when the number of teaching physicians was large. In addition, the magnitude of elasticity of the interaction and the number of teaching physicians were larger in large cities than those in small cities, suggesting that the antagonistic effect of the number of teaching physicians is larger in the urban setting. These results suggest that an increase in the number of teaching physicians may be a more effective strategy to attract residents in the urban setting than an increase in salary, while the increase in the number of teaching physicians may be a difficult alternative under the current shortage of physicians, especially in rural areas . Offering a higher salary for young residents may be more feasible instead. Some previous studies argued that financial incentives alone are not sufficient, and that the strategy for staffing remote rural areas should be multi-facetted and comprehensive [6, 8, 13]. Our results rather indicated that resource could be strategically allocated between monetary incentives and training environments, according to local conditions.
Our study has several limitations. First, almost half of the hospitals included in the national residency programme were excluded from our sample because of the limitation in data availability. Because the submission of hospital information was voluntary, hospitals with a small-scale programme tended to be excluded from the Guidebook. We compared all the hospitals included in the Guidebook (n = 705 in 2006, and 683 in 2009) with those included in our panel analysis. We found that the number of teaching physicians in 2006 was smaller in all the hospitals included in the Guidebook than those included in our panel analysis (10.11 persons versus 10.40 persons, P = 0.04), and the number of inpatients was also smaller in 2006 and 2009 (in 2006, 359 versus 375 patients, P = 0.03; in 2009, 332 versus 348 patients, P = 0.02). Otherwise, the characteristics were comparable between the two datasets. In our study, the descriptive statistics on the number of hospital beds or doctors were similar to those of Kawamura’s study, which had a cross-sectional and larger sample than that in our study . However, in 2009 in the university-affiliated hospitals, the average number of beds was 596, the average number of outpatients was 977 per day, and the average number of doctors was 47 per 100 hospital beds ; these numbers were much larger than in our sample. Therefore, we acknowledge that our estimation of the effect of salary in non-university hospitals may not necessarily hold true in the university hospitals. We also cannot deny the possibility that our results may have limited generalizability, and these should be confirmed with longer-term and larger panel data from teaching hospitals in the country.
Second, the natural change in residents’ preferences over the studied period could have affected our results. For example, the number of female physicians has been continuously increasing to more than 30% of newly qualified physicians . Sex is significantly related to a resident’s choice of a university hospital as a training site , and also with the likelihood of satisfaction with salary . Because the number of residents by sex was not available, we could not account for the effect of the resident’s sex in our analysis. The results of this study should be confirmed with sex-specific datasets, or more preferably, with microdata of residents’ preferences.