Types Multinomial Logit Analysis

If HRM practices lower moral hazard behavior, HRM practices ought to have an impact on the type of claims that are filed. If the potential for moral hazard is greatest where the costs of monitoring are largest, then relatively more evidence of moral hazard should be seen with respect to those types of claims that are most difficult to monitor. Again, this is especially true for sprains, strains, and cumulative trauma conditions. 

Butler, Durbin, and Helvacian (1996) use this distinction between diffieult-to-monitor and easy-to-monitor injuries to explore whether soft-tissue injury elaims correlate with level of benefits and spread of HMOs. They find in their 10-year, 15-state sample of workers compensation claims that the proportion of claims attributable to soft-tissue injuries rose from 44.7 percent of all claims in 1980 to 50.6 percent in 1989. Concurrently, the share of costs attributable to soft-tissue injuries rose from 41 pereent to 48.8 percent. The share of costs for injuries that crush or fracture a bone—easy-to-monitor claims—is the only category that declined between 1980 and 1989. Using a multinomial logit model, the authors determine that most of the increase in soft-tissue injury is attributable to the expansion of HMOs. Specifically, they ascribe the rise in such injuries to moral hazard response by HMO providers, who increase their revenue by classifying as woik-related injuries as many health conditions as possible.  

We build on the work done by Butler, Durbin, and Helvacian (1996) by considering whether HRM practices affect the distribution of injuries. Given prior evidence on soft-tissue sprain and strain, we would expect to see additional HRM practices associated with fewer sprains and strains (particularly back sprains and strains), and with relatively more fractures and lacerations, //HRM practices are reducing woik-place injuries through a claims-reporting response. 

On the other hand, if HRM practices work to reduce claims-report-ing moral hazard, we would expect to find them reducing the proportion of back sprains and strains and increasing the relative number of fractures, contusions, and cuts. 

To place the ideas discussed above into a statistical framework, we assume the typical worker experiences one of five states. The worker may not have any type of health impairment whatsoever, or the worker may experience some sort of injury that places him into one of four injury categories 1) fractures, contusions, and cuts 2) back sprains and strains 3) nonback sprains and strains and 4) all others. 

---