Nurse Staffi ng Results

V. Results

Below, I present estimates of the impact of the minimum staffi ng leg- islation on staffi ng levels, quality of care measures, and a variety of other margins for skilled nursing facilities in California. In each section I present graphical evidence allowing both for visual identifi cation of the impact of the legislation and evaluation of the key identifying assumptions, followed by OLS estimates of Equation 1 to provide point estimates and standard errors.

A. Nurse Staffi ng

Figure 1 provides the most direct illustration of the effect of the minimum staffi ng requirements on overall staffi ng levels, and depicts the essence of the research de- sign employed in the paper. I divided fi rms with initial staffi ng levels below the 3.2 threshold into seven roughly equal groups based on their average HPRD staffi ng level in the years 1997 and 1998. These seven “decile groups” all have roughly 100 fi rms between 99 and 101, with decile 1 having the 100 fi rms with the lowest initial staff- ing levels, decile 2 having the 100 next lowest, and so on. Similarly, I divided fi rms above the 3.2 threshold into three roughly equal groups, so deciles 8 through 10 each contain roughly 90 fi rms with progressively higher pre- period staffi ng levels. The six lines in the fi gure all depict trends in the average staffi ng levels between 1996 and 2004 for six of these decile groups; the remaining four decile groups are omitted to simplify presentation of the results. Several features of this fi gure are important to note. The most striking feature is the trend break in average staffi ng levels that oc- curs in the year 2000, when the staffi ng legislation becomes effective. As we would expect if this change is due to fi rms attempting to comply with the law, the increases in staffi ng are more pronounced for fi rms with lower initial staffi ng levels, and are smaller for the fi rms in the top three staffi ng deciles who already were in compliance with the law as of 1997 and 1998. Staffi ng levels are mostly unchanged throughout the period for fi rms in the 10th decile—with the highest initial staffi ng levels—but increase by nearly 0.6 HPRD for fi rms in decile 1, with intermediate changes for the deciles in between. If nurse staffi ng levels positively affects quality of care in nursing homes, then we should expect to see similar patterns in the changes of such outcomes: Firms should see improvements in quality that are proportional to how far below the threshold their staffi ng levels were in the preperiod that is, proportional to GAP, and fi rms initially in compliance should see relatively small changes with little relationship to initial staffi ng levels. In this research design, drawing a causal link between staffi ng changes and quality of care changes rests on an assumption that in the absence of the minimum staffi ng law, the trends in unobserved determinants of outcomes should be similar for fi rms with different initial staffi ng levels. The fact that Figure 1 shows staff- ing levels trending in similar ways in the four years before AB 1107 becomes effective lends support to this assumption. 16 16. The changes between 1998 and 1999 apparent in the Figure are likely due to mean reversion. I measure initial staffi ng levels with a two- year average and use 1999 as the base year in regression estimates of equa- tion 1 to mitigate division bias resulting from measurement error. Figure 2 displays the impact of the law on the staffi ng levels of particular types of nurses—nurse aides top panel and registered nurses bottom panel—in a way that is more tightly linked to the research design described in Equation 1. Each dot in the fi gures represents one of the nursing homes in the sample, and plots the change in the log of total annual hours worked for each type of nurse from 1999 to 2004 against its average staffi ng level in 1997 and 1998, the two years prior to passage of the staffi ng law. The dashed line in the Figure shows fi tted values from the simplest specifi cation for Equation 1, regressing the change in staffi ng levels on a constant term and GAP. For nurse aides, the estimated coeffi cient on GAP is 0.23 standard error: 0.029, indicating that on average fi rms increased their nurse aide staffi ng by about 23 percent for every 1 HPRD below the 3.2 threshold they were in the preperiod. While the estimated constant term is nonzero, indicating that fi rms initially above the threshold did increase their staffi ng levels, there does not appear to be a relationship between initial staffi ng levels and the size of this increase for fi rms that were already in compliance. This is borne out by the solid line in the Figure, which shows that a Figure 1 Effects of Legislation on Nurse Hours per Resident Day by Initial Staffi ng Level “Deciles” Notes: HPRD are measured with no double counting of licensed nurses, according to the rules in place after 2000. The fi gure displays trends in the relevant outcome for each of six “decile groups.” The line marked “1” displays the average log annual hours worked in the 100 fi rms with the lowest staffi ng level in 1997 and 1998 measured by average HPRD over those two years whereas the line marked with “10” depicts the same outcome for the 90 fi rms with the highest staffi ng levels in 1997 and 1998. 1 1 1 1 1 1 1 1 1 3 3 3 3 3 3 3 3 3 5 5 5 5 5 5 5 5 5 8 8 8 8 8 8 8 8 8 9 9 9 9 9 9 9 9 9 10 10 10 10 10 10 10 10 10 2.5 3 3.5 4 4.5 Hours per Resident Day 1996 1998 2000 2002 2004 Year nonparametric local linear regression nearly matches the fi t of the linear specifi cation in Equation 1. 17 While the bottom panel shows a similar pattern of results for regis- tered nurses—the estimated coeffi cient on GAP is 0.19—the wider dispersion of dots around that trend line suggests that the result is less statistically precise. Similar graphs for LVNs show average employment levels increasing by roughly the same magnitude across all levels of preperiod staffi ng. Table 2 adds some precision to these fi ndings, showing the results of estimating various specifi cations of Equation 1 for nurse aides, licensed vocational nurses, and registered nurses. Three models are estimated for each type of nurse using the change 17. The solid line shows the connected predicted values from a local linear regression using a triangular kernel and a bandwidth of 0.05 HPRD, shown over the subset of observations with 2.5 to 4 HPRD on the x- axis. Nursing Assistants Facilities with extreme values are omitted for presentation purposes. Data for 913 of 961 facilities shown in fi gure. Figure 2 Effect of Legislation on Nurse Staffi ng by Occupation Notes: Each fi gure displays the change in log of total annual hours worked for the relevant occupation as a function of preperiod staffi ng level for each nursing home in the sample. The dashed line plots the fi tted values from a regression of this change on GAP and a constant term as described in the text, and the solid line shows the fi t of a local linear regression through these points using a triangular kernel and bandwidth of 0.05 HPRD. Coeff on GAP: .232 .0289 1999 Average: 10.978 -.5 .5 1 Change: 1999 to 2004 2 2.5 3 3.5 4 4.5 1997-98 Ave. HPRD in log annual hours between 1999 and 2004 as the dependent variable: the fi rst in Col- umns 1, 4, and 7 regresses the change in outcomes on GAP and a constant term only, the second Columns 2, 5, and 8 adds county fi xed effects, and the third adds a linear control for average HPRD in 1997 and 1998 recentered so that 3.2 HPRD is zero. As described above, the causal effect of the staffi ng law is captured by the coeffi cient on GAP. For nurse aides, the estimated coeffi cient is positive and signifi cant across all three models: The law caused nursing facilities to hire more nurse aides in proportion to how far out of compliance they were before the staffi ng law was passed. To get a sense for the magnitude of the coeffi cients, note that fi rms with initial staffi ng levels below the 3.2 threshold had an average staffi ng level of 2.7 HPRD, or a GAP of 0.5. Thus, for the average fi rm in this sense not already in compliance, the law induced a roughly 10.4 percent increase after four years. 18 For more skilled nurses, the results are quite different. On average, nursing homes increased their employment of LVNs by about 0.20 log points from 1999 to 2004 the constant term in Column 4 of Table 2. But this increase was uniform across 18. The calculation is 0.5 × 0.2082, taking the coeffi cient from Model 2. The estimate is smaller, and some- what less precise, when I control additionally for HPRD in the regressions Column 3. Figure 2 continued Registered Nurses Facilities with extreme values are omitted for presentation purposes. Data for 902 of 950 facilities shown in fi gure. Coeff on GAP: .1976 .0927 1999 Average: 8.923 -3 -2 -1 1 2 Change: 1999 to 2004 2 2.5 3 3.5 4 4.5 1997-98 Ave. HPRD T he J ourna l of H um an Re sourc es 48 Table 2 Effects of Staffi ng Legislation on Nurse Employment change in log annual hours between 1999 and 2004 Nurse Aids LVNs RNs 1 2 3 4 5 6 7 8 9 GAP 0.232 0.2082 0.1352 –0.001 –0.0095 0.0628 0.1976 0.1751 –0.2314 [0.0289] [0.0289] [0.0446] [0.0468] [0.0494] [0.0826] [0.0927] [0.098] [0.2000] HPRD 9798 – 3.2 –0.0468 0.0465 –0.2588 [0.0213] [0.0438] [0.1222] Constant 0.0916 0.1979 –0.0773 [0.0114] [0.0188] [0.0257] County fi xed- effects No Yes Yes No Yes Yes No Yes Yes R 2 0.081 0.120 0.125 0.000 0.047 0.049 0.006 0.052 0.052 Observations 961 961 961 958 958 958 952 952 952 Notes: The table presents estimates of several specifi cations of Equation 1, described in the text. In each column the dependent variable is the change in log total annual hours worked between 1999 and 2004 of the relevant nurse occupation. The key independent variable of interest is GAP, which represents the gap between the 3.2 HPRD minimum staffi ng level and a facility’s pre- period 1997–98 staffi ng level, defi ned to be 0 for fi rms with staffi ng levels already above 3.2 in the pre- period. HPRD 9798 – 3.2 is a linear control for preperiod staffi ng levels, “recentered” so that 3.2 is equal to zero. Columns 2, 3, 4, 6, 8 and 9 also include a vector of 31 county fi xed effects, which control for county level changes in the determinants of staffi ng levels such as local labor supply and Medicaid Medicare reimbursement levels. Robust standard errors are presented in parentheses. homes with different initial staffi ng levels, and in fact the increase is fi rst discernible between 1998 and 1999, before the staffi ng legislation was passed. 19 For RNs, there is an evident reduction in the average hours worked across all fi rms, but the reduc- tions are largest among those fi rms that had the highest initial staffi ng levels. Thus, the coeffi cients on GAP in Table 2 in Models 1 and 2 suggest that the law may have increased RN staffi ng by about 10.5 percent for the homes initially out of compli- ance with the law but this should be understood as the law preventing a decline in these hours in an absolute sense. But, these results are not robust to the inclusion of a control for HPRD, and an analysis of preperiod trends shows that the change had already become apparent before the law had passed. Other research has suggested that a switch to Medicare prospective payment for skilled nursing homes in the Balanced Budget Act of 1997 may have created fi nancial pressure beginning in July of 1998 for nursing homes to substitute away from RNs in favor of LVNs Konetzka, Norton, Sloane, Kilpatrick, and Stearns 2006. 20 It may be that fi rms with low staffi ng levels were more constrained in their ability to do this, producing the patterns in staffi ng by occupation observed.

B. Patient Outcomes and Facility Defi ciencies