The Labor Market Effects of Minimum Staffing Legislation:

Transcription

1 The Labor Market Effects of Minimum Staffing Legislation: Direct Tests for Monopsony in High- and Low-Skilled Labor Markets Jordan D. Matsudaira 1 Cornell University Policy Analysis and Management March 2008 Very Preliminary: please do not cite or circulate without permission 1 I thank and hold blameless John DiNardo for helpful discussions and advice. I am grateful to the Robert Wood Johnson Foundation for financial support. Comments are welcome at jdm296@cornell.edu.

2 Abstract In , California passed legislation requiring all licensed nursing homes to maintain a minimum number of hours of nurses per patient day aimed at improving quality of care. The law created pressure for firms below the mandated threshold of 3.2 hours per patient day (hprd) to increase nurse staffing levels, but allowed them some freedom over what types of nurses could be hired to comply with the law. This paper makes two contributions. First, it documents that the law was highly effective in increasing overall hours per patient day in facilities that had been below the mandated threshold before the legislation passed. On the other hand, nursing homes comply with the law primarily by hiring less skilled nursing assistants, significantly reducing the fraction of total nursing hours provided by licensed nurses. The second contribution of the paper addresses a long-standing debate over nursing labor, concerning the degree of monopsony present in the market. Using the mandated changes in staffing levels as a firm specific instrument for changes in the quantity of hired labor, the paper directly estimates the inverse elasticity of the labor supply curve for various types of nursing labor. The results suggest negligible monopsony power in the market for nursing assistants a result at odds with some recent other findings but further work seems necessary to reliably examine the case of the registered nurses that are more commonly at the center of monopsony discussions.

3 The Labor Market Effects of Minimum Staffing Legislation Direct Tests for Monopsony in High- and Low-Skilled Labor Markets Preliminary: Please do not cite or circulate without permission. 1 Introduction If an employer needs to hire more workers, must it raise the wage it offers in order to do so? The answer to this seemingly innocuous question distinguishes between two competing theories of how labor markets function. Most of labor economics is built on the assumption of perfect competition in the labor market: employers can hire as many workers in a given class as they desire at a constant wage set by the market. In other words, the elasticity of the labor supply curve to an individual firm is infinite. 1 Recently, however, there has been renewed interest in economics in the relevance of monopsony models that begin from the premise that the labor supply curves to individual firms are upward sloping. Which model more accurately describes the labor market has broad reaching implications for many areas of public policy concern, such as the desirability of minimum wage floors, the sources of race and gender pay gaps, and the causes of increased wage inequality to name only several. In contrast to other areas of economics, monopsony has long been a fixture in discussions of nursing labor markets. At least since Yett (1975), imperfect competition has been proposed as a primary driver of a perceived shortage of nurses. Analysts have pointed to high degrees of concentration, and limited geographic and occupational mobility options for nurses as reasons why the adjustment processes assumed to underly a perfectly elastic labor supply curve to the firm might fail. Despite the status of nurse labor as one of the textbook examples of monopsony, however, few empirical studies have convincingly demonstrated the presence of monopsony in nursing or any other market. The reason for this, as noted by Manning (2003), is that it is difficult to find firm specific instruments for either wages or employment levels that can allow for consistent estimation of the slope of the labor supply 1 Alan Manning, in his canonical text on monopsony (2003), asks the question this way: What happens if an employer cuts the wage it pays its workers by one cent? Under perfect competition, all workers quit immediately. 1

4 curve to individual firms. In this paper, I attempt to directly estimate the degree of monopsony power for various types of nursing labor by examining the labor market effects of a recent policy change affecting long-term care facilities in California. In 1999, the California assembly passed a law 2 requiring that all long-term care facilities employ a minimum number of nursing hours for each resident day. The law required that all facilities maintain at least 3.2 hours of nursing staff per resident per day (hprd), and established a range of penalties for noncompliance. In essence, the law creates pressure on firms who were below the 3.2 hprd standard before the law passed to increase their staff. The crux of the research design in this paper is thus to estimate the effect of increasing the number of hours of various nursing occupations on their wages, using the gap between a firm s initial staffing level and the 3.2 threshold as an instrument for its subsequent change in nursing hours. To the extent that initial staffing levels are uncorrelated with changes in other factors affecting labor supply, the approach will yield consistent estimates of the inverse elasticity of labor supply. If monopsony is a significant feature of the labor market for nurses, this elasticity should be positive. A novel contribution of the paper is to attempt a comparison of monopsony power among higher skilled registered nurses and less-skilled nurses aides. While the former has been the subject of most discussions on monopsony, Card and Krueger (1995) touched off a controversial debate among labor economists by suggesting market power might be an important feature of the low-skilled labor market. This paper attempts to assess the relative amounts of monopsony power for these groups of high and low skilled nurses working for a common set of employers. Below, the paper shows that the minimum staffing legislation was quite effective in increasing staffing levels in long-term care facilities that had been below the 3.2 hprd threshold prior to the passage of the law. For nursing homes that had been in the lowest quintile of hprd in 1999, overall hours increased by about 30 percent between 1999 and The 2 The law was Assembly Bill 1107, passed in 1999 and taking effect in January of

5 increase, however, was not uniform across nursing occupations. The law had its greatest impact on the number of nursing assistants hired in facilities with low initial staffing levels. Moreover, in a period where facilities were substituting away from more highly skilled registered nurses, the law prevented these declines in facilities with low initial staffing levels. Overall, then, the law improved staffing levels but the fraction of hours provided by more skilled nurses declined. If the labor market for nurses is monopsonistic, these large increases in staffing should have been accompanied by increased wages. There does not appear, however, to have been a causal impact of the legislation on wage levels for any category of nursing labor. At first blush, then, the results suggest that perfect competition is the best model for the labor market for nurses. One strong caveat to this result is that due to data limitations I have not yet accounted for competitive dynamics among potential employers that are likely an important consideration in the market for more highly skilled nursing labor. This weakness notwithstanding, the findings of no monopsony in the market for less skilled nursing assistants adds interesting texture to a growing body of research on the empirical relevance of monopsony. The remainder of the paper is organized as follows. The next section briefly discusses theories of monopsony and its sources, as well as some empirical work on monopsony, most of which conveniently to focuses on nursing markets. Section 3 then presents a simple empirical model of the labor supply of nurses to firms and describes how the minimum staffing legislation for California nursing homes can be used to estimate the inverse elasticity of the supply curve. After a brief description of the data sources used in the paper and a statistical overview of the nursing homes included in the analysis, I turn to the empirical results in Section 5. The first part of the section describes the effects of minimum staffing legislation on the level and composition of nursing labor in nursing homes, itself an important policy issue. I then turn to assessing impacts of the law on nurse wages and present tests of the relevance of monopsony based on estimates of the inverse elasticity of supply. Section 6 discusses the results in the context of other work on monopsony, and the final 3

6 section concludes. 2 Background and previous research In her book The Economics of Imperfect Competition (1933), Joan Robinson discussed two broad scenarios under which employers might have market power over their workers. If either an employer were so large as to provide the only source of jobs for a class of worker, or a group of employers were able to collude in setting wages then they would optimize by considering the increased wages necessary to draw workers into the market (either from other occupations or other geographic areas). For example, a monopsonistic organization might hire workers to maximize joint profits by solving the problem max L π(l) = R(L) w(l)l, where L is the amount of labor hired, R(L) is a joint revenue function, and w(l) is the labor supply function to the market. The first order conditions of this problem imply (1) MRP w w = w w L L ɛ where MRP is the marginal revenue product of labor, and ɛ is the inverse elasticity of labor supply. In this setting, wages would be set below the marginal revenue product of workers in proportion to the (inverse) elasticity of the supply of labor to the market. Employment would also be set below the competitive level, and there would be an excess demand for workers at the prevailing wage. Archibald (1954) labelled the gap between the number of employees that would be hired if labor supply were perfectly elastic at the going wage and the number actually hired at a firm as the firm s vacancies. Based implicitly on logic embedded in models of this sort, a number of studies have investigated the extent of monopsony power in the labor market by looking for correlations between firm concentration ratios and wage levels. 3 3 Boal and Ransom (1997) show that in a Cournot model of oligopsony, an employment weighted average of rates of exploitation can be expressed as E = P h n i Ei L i = ɛ Pn ` Li 2i L i, where L L i is employment at firm i, and ɛ is the inverse elasticity of labor supply. The term in brackets is the Herfindahl index of employment concentration that is sometimes used in empirical work. 4

7 Many studies of monopsony have examined the market for registered nurses (RNs) which has long been imprecated as a leading example of a monopsonistic labor market. Several facts make nursing a natural candidate: there tend to be relatively few employers of RNs predominantly hospitals in any given geographic area, particularly in rural areas: most hospitals are the only hospital in their community. Nurses are also often secondary wage earners in a family, and so may be relatively unlikely to move to a new geographic area in response to higher wage offers. Moreover, RNs have high occupation-specific skills and so are unlikely to easily switch occupations. Further, there has been a purported shortage of nurses, manifest by vacancies reported by hospitals, since the early part of the 20th century. 4 Despite this strong prima facie case 5 and early results consistent with the presence of monopsony, the evidence from studies correlating concentration ratios with wages is not convincing. Several early studies found cross-sectional correlations between concentration ratios and nurse wage levels: for example, Hurd (1973), Link and Landon (1975), and Feldman and Scheffler (1982). This finding is evidence for monopsony only if the other factors related to labor supply and demand are held constant across markets (Boal and Ransom (1997)). Subsequent studies, such as Adamache and Sloan (1982) and Hirsch and Schumacher (1995), have found that controlling for such factors e.g., population density or wages in other occupations eliminates the relationship between nurse wages and hospital concentration. 6 Monopsony theories based on large firms or groups of firms monopolizing employment were initially dismissed by many economists because examples of company towns were rare 7 and the incentives to break collusive agreements seemed too great for them to be maintained. Robinson, however, argued that monopsony-like conditions could arise in a labor market without large firms if the supply curve to individual firms was upward sloping. This could 4 See, for example, Yett (1975). 5 See, however, Rosen (1970), who was an early skeptic about monopsony in nursing. 6 Studies of other labor markets have yielded mixed results. See Boal (1995) for a study of coal miners in the early 1900s, or Luizer and Thornton (1986) for a study of school teachers. 7 Bunting (1962), for example, showed that it was rare for the biggest firms in an area to employ a large fraction workers in a local labor market. 5

8 happen because there may be a certain number of workers in the immediate neighbourhood and to attract those from further afield it may be necessary to pay a wage equal to what they can earn near home plus their fares to and fro; or there may be workers attached to the firm by preference or custom and to attract others it may be necessary to pay a higher wage. Or ignorance may prevent workers from moving from one firm to another in response to differences in the wages offered by the different firms (Robinson 1933, 296). This observation appears to have been initially well received by some labor economists. Writing in 1946, Lloyd Reynolds predicted The view that labor-market imperfections result in a forward-rising supply curve of labor to the firm... first elaborated by Mrs. Robinson... seems well on the way to being generally accepted as a substitute for the horizontal supply curve of earlier days (Reynolds 1946, 390). Despite Reynolds enthusiasm, however, the ensuing sixty years produced very little direct evidence on the elasticity of the labor supply curve to individual firms. The reasons for this are readily apparent: the observed changes in wage and employment levels across firms are in general the result of changes to both supply and demand. So long as the supply curve is shifting, the observed equilibria can not reliably be used to identify its slope. As Manning (2003) summarizes the state of literature: This is all rather depressing: a good estimate of the elasticity of the labor supply curve facing the firm seems very elusive so perhaps there is a very good reason for the lack of research into this area. Progress seems to be dependent on finding a good firm-level instrument (96). Only a handful of studies have tried to use firm-level instruments to estimate the labor supply elasticity to the firm. Sullivan (1989) estimates the elasticity of labor supply to individual hospitals using data from 1979 to He derives labor supply equations based on different assumptions about the nature of competition among firms, and controls for hospital and region specific fixed-effects to estimate the (inverse) elasticity of supply. 8 The simultaneity problem is addressed by using the number of caseloads and average length of 8 For example, assuming Nash equilibrium in employment levels his estimating equation is w rit = α i + δ rt + θn rit + γ(on) rit + ɛ rit, where w rit represents log wages for hospital i in region r at time t; n rit is the log number of nurses employed; and (on) rit the the log of the sum of nurses employed at other hospitals. θ represents the inverse elasticity of supply. 6

9 stay as instrumental variables for the number of nurses at hospitals. While the validity of this instrumental variable has been questioned 9, Sullivan finds that the inverse elasticity of supply is about.79 (with standard error of.13) over a 1 year period and.26 (.07) over a three year period and asserts this represents a significant amount of market power for hospitals. 10 Instrumental variable strategies are also employed by Staiger et al. (1999) and Falch (2003) who use legislated wage changes in Veteran s Affairs (VA) hospitals and public schools in Norway to identify the labor supply elasticity of RNs and teachers, respectively, to firms. Similar to the analysis in Sullivan (1989), Staiger and his coauthors adopt an explicit model of oligopsony where hospitals compete most intensively with hospitals in close proximity. They demonstrate that VA wage changes have effects on the wage levels of RNs at nearby hospitals (up to 30 miles away), suggesting that hospitals do have the ability to set wages. Using gaps between the newly legislated wage and wages at the time of the legislation as instruments for wage changes, the estimated labor supply elasticity over a two year period ranges from near 0 to 0.2 with standard errors close to 0.13 (or an inverse elasticity ranging from about 5 to infinity). Even using the upper bound of the 95 percent confidence interval from Staiger et al. s estimates implies that the inverse elasticity of labor supply is at least 2: far from the zero assumed by the theory of perfectly competitive labor markets. Several relatively recent studies have found indirect evidence for monopsony, and this evidence suggests that the phenomenon may not be limited to a subset of relatively skilled workers in industries where employment tends to be concentrated in large employers. The highest profile amongst these studies is surely Card and Krueger (1995) s survey of research 9 Staiger, Spetz and Phibbs (1999) point out that Sullivan s sample brackets a period when Medicare s Prospective Payment System is introduced, and suggests that much of the variation in hospital days over the period was therefore endogenous. Manning (2003) suggests that caseloads might be related to population shocks, and thus might fail the exclusion restriction. 10 In a static model these elasticities can be used to compute the markup of marginal product over wages using equation (1). In this case, the estimates would imply wages are between 43 percent (for one year changes) and 21 percent (for three year changes) below marginal product. In a dynamic setting, however, this rate of exploitation is a weighted average of short and long run elasticities where the weights are a function of a firms discount rate. Assuming a long run elasticity of zero, Boal and Ransom (1997, 105) suggest Sullivan s estimates imply that wages might be set between 87 and 96 percent of marginal product. Using this logic, they characterize Sullivan s results as being suggestive of only slight market power for hospitals. 7

10 showing negligible or even positive employment effects of the minimum wage. Card and Krueger suggest that monopsony models are better able to reconcile the results of minimum wage studies than perfect competition models, and suggest that monopsony models based on search costs might best describe the labor market dynamics of low-skilled workers in the fast-food industry. This claim has been quite controversial, as many economists dismiss out of hand the notion that important frictions exist for low-skilled workers. This paper hopes to contribute to the existing body of literature on monopsony in two key ways. First, I will directly estimating the elasticity of labor supply to individual establishments using a new plausibly exogenous instrument for changes in employment at the firm level. And second, by exploring the degree of monopsony power in the market for nurse aides, licensed vocational nurses, and registered nurses, I hope to contribute to the debate on whether monopsony power in the market for low skilled workers might be as important as in the market for higher skilled workers. 3 Identification Strategy and Minimum Staffing Ratio Legislation As outlined above, theoretical models of monopsony suggest a general model for labor supply to the individual firm can be written as w i = f(n i, X i, X i ), where w i is a firm s wage; n i is its employment of nurses; X i is a vector of firm and local market characteristics; and X i represents the actions (e.g. wages or employment levels) of competitor firms. It is important to note that the source of monopsony power has important implications for specifying the labor supply equation. For example, if the dominant forces in wage determination arise from oligopsonistic competition for workers among differentiated firms, then understanding which firms compete with each other and the nature of this is important. On the other hand, if search frictions are the prime cause of monopsony power, then worrying about the actions of other firms may be less important There is scant evidence on this. Sullivan (1989) finds little impact of competitors actions on wages in the short-run, but some impact over a three-year period. Staiger et al. (1999) estimates of elasticities do not 8

11 In this paper, I adopt a very simple model of labor supply to the firm for labor type k: (2) w k irt = β k 0 + β k 1 n k irt + α k i + θ k r t + ɛ k irt. where w k irt represents real hourly wages for workers in the kth occupation at facility i in region r year t; n irt represents total hours worked, and ɛ k irt represents other supply and demand shocks affecting wages and employment. The equation will be estimated separately for RNs, LPNs, Nurses Aides, and other non-nursing labor groups. α i represents fixed firm and labor market characteristics affecting the desirability of employment, like proximity to population centers. These effects are controlled for by estimating the equation above in differences of different lengths (one to four years), which also allows me to estimate the elasticity of supply in both the short and slightly more medium run. It would seem natural, as Sullivan (1989) finds, that the labor supply curve becomes more elastic over longer windows as more adjustment can occur. Finally, θr k t represents a region specific time trend, that is meant to absorb changes in local labor market conditions such as wages at competitor firms. This will approximately control for changes in X to the extent that firms are small relative to the market (so X is the same for all firms). Future work will explore the consequences of controlling separately for the wages and employment levels of geographically close competitors as suggested by models of geographic differentiation. 12 With data differenced over d years and region specific time trends, the equation above becomes (3) d w k ir = β k 1 d n k ir + θ k r + d ɛ k irt. In the equation above, β k 1 represents the inverse elasticity of supply, and a test of β k 1 = 0 amounts to a test of the null hypothesis of a perfectly competitive labor market for seem heavily influenced by the inclusion of competitors wages. 12 These considerations seem more important for labor groups with high occupation-specific human capital such as RNs or LPNs. For nurse aids the set of likely alternative employers probably includes many firms employing low-skilled labor outside of the health care industry, and viewing a particular LTC employer as atomistic may not be unreasonable. 9

12 occupation k. The primary empirical challenge involved in in consistently estimating β 1 is that d ɛ ir may be correlated with d n ir for firms in the same region. In particular, so long as the firm specific demand-side shocks leading to employment changes are correlated with supply-side shocks (that is, ɛ irt ) then estimates of the inverse elasticity will be inconsistent. 13 In this paper, I argue that the passage of minimum staffing legislation in California provides a new opportunity to overcome this basic simultaneity problem and identify the elasticity of nursing labor supply to individual facilities. In 1999, the California legislature responded to concerns over the quality of care in nursing homes and hospitals by passing legislation requiring minimum nurse staffing levels. Governor Gray Davis signed California Assembly Bill 1107 addressing nursing homes into law on July 22, 1999 and the law became effective on January 1, A separate law, AB 394, addressed hospitals but was not implemented until AB 1107 required that all skilled nursing and intermediate care facilities provide a minimum of 3.2 nursing hours per resident day (hprd) and established a range of penalties for noncompliance. The law thus put pressure on all firms with staffing levels below 3.2 hprd in 1999 to increase their staffing levels. More specifically, the law should have had no effect on the nursing hours employed by firms that were above 3.2 hprd in 1999, and should have increased hours worked for firms below the threshold in proportion to how far they were below the standard when the law passed. Based on this logic, I define a variable GAP i equal to the absolute difference between facility i s 1999 staffing level (in hprd) and the 3.2 threshold if the firm is below the threshold, and equal to zero if the firm is above the threshold. To be clear, GAP is positive for all firms below the threshold, and greatest for firms that were furthest below the legislated standard. I then estimate equation (3) above via two-stage least squares, using GAP i an instrument for d n k ir. So long as the unobserved determinants of wages aren t changing in a way that is correlated with a firm s 1999 staffing level, then this approach should allow for identification of β1 k. In most analyses, I present estimates from a more flexible model that allows changes in the outcomes to depend on a linear function of the 1999 staffing level 13 See Manning (2003, 83) for an illustration in a simple model. 14 Facilities were notified, however, that enforcement of the new standard would begin in April of as 10

13 as well as GAP. In this case, the coefficient on GAP is identified by the difference in this relationship for firms below the threshold (affected by the law) and firms above the threshold. 15 In the results section below, I present estimates of the impact of the staffing legislation on staffing levels and then on wages, and then use these results to estimate the elasticity of the supply curve. I present results separately for several different nursing occupations: managers, registered nurses (RNs), licensed vocational nurses (LVNs), and nursing assistants (NAs). Each of these occupations may have been affected by the law, but due to differences in non-nursing alternative employment opportunities we may expect the extent of monopsony power of firms over each of these occupation groups to differ. I also present estimates of the staffing, wage effects, and elasticity estimates over different time spans to investigate whether longer run elasticity estimates spanning up to four years appear to be higher (or lower for the inverse elasticity) than do short run estimates measuring wage responses over only one or two years. Before turning to the results, I briefly describe the data sources and provide an overview of the sample of nursing homes and nurse characteristics. 4 Data and Summary Statistics All data on nursing home characteristics, including the total number of hours worked and salary and wage information by nurse occupation group comes from the Long-Term Care Facilities Annual Financial Data collected by California s Office of Statewide Health Planning and Development. These data consist of financial, utilization, and staffing information contained in Long-term Care Facility Integrated Disclosure and Medi-Cal Cost Reports (known as Disclosure Reports ) submitted by long-term care facilities and audited by staff at OSHPD. This paper uses publicly available data for nearly all licensed long-term care facilities operating in California between 1996 and To form an analysis sample, I 15 In what follows, I use GAP i as the instrument for the change for each type of nurse labor. This is not strictly kosher in the model of hiring left implicit here each type of labor is endogenously determined. More instruments and/or more structure are necessary to identify all the elasticity parameters of this model. This will be addressed in future work. 11

14 start with the 1,215 facilities that submitted data to OSHPD in 1999, the year just before minimum staffing legislation passed. I drop 24 facilities whose data are deemed noncomparable by OSHPD these are primarily congregate living health facilities or hospices (22 of the 24), and homes operated by either charities or government agencies. I further drop 14 facilities whose measured hprd values in 1999 seem implausibly high (greater than 6) or low (less than 1). From the 1,177 facilities that remain, I use a subset of 1,134 firms for which data is available in every year from 1999 through Appendix Table 1 shows how this sample restriction is related to 1999 hprd levels. There are 43 facilities present in 1999 that exit the sample by 2003, but this attrition appears to be unrelated to hprd in 1999 suggesting that survivorship bias may not be important. In some analyses I further restrict the sample to facilities with valid data in the three years (from 1996) leading up to 1999, resulting in 23 additional firms being dropped from the sample. This restriction drops a slightly higher fraction of firms from the highest 1999 hprd category, but the magnitude of any bias arising from this restriction is likely to be small. Table 1 presents descriptive statistics for the 1,134 firms used in the main set of analyses, broken down by quartiles of the 1999 staffing distribution. The facilities in the sample range in size from a low of 19 beds to 391 beds at the largest facility with a mean and median of 98. Firms employed an average of about 97 workers with the firm at the 10th percentile of size employing 47 workers and the firm at the 90th employing 160 workers. Total health care revenues averaged about $4 million. All of these measures of firm size suggest that firms in the lowest quartile were smaller, but that the differences among firms in the top three quartiles in terms of size were relatively small. Private (investor-owned) firms dominate the nursing home market, accounting for about 87% of homes. The remainder are run by non-profit organizations (8 percent) and churches (5 percent), with a handful of firms run by the federal (VA) or county governments. There is a clear pattern between staffing levels and ownership: a heavy majority of all non-profit and church run facilities are in the highest staffing quartile. Occupancy, measured as the total number of patient days divided by the number of 12

15 available bed-days was uniformly high with a median of 90 percent the firm at the 10th percentile had an occupancy rate of 74 percent. Most patient days 63 percent were paid for by MediCal 16, but 25 percent of patient days were paid for by individuals (presumably out of pocket), 6 percent from Medicare, with the remainder coming from private insurance. Here again there are pronounced differences related to staffing levels. In particular, firms with higher staffing ratios are much more likely to have patients paying out of pocket rather than using Medical, suggesting that they serve a higher income clientele. Nearly all of the facilities were skilled-nursing facilities with only about 3 percent comprised of intermediate care facilities for patients with less acute needs. As can be seen in Figure 1, the facilities were concentrated near population centers in Los Angeles, the Bay Area, and Sacramento, but were generally spread throughout the state. Appendix Table 2 shows the number of facilities in each of 31 county groups, formed by combining counties with few homes out of California s 58 counties. In 1999, only 6 counties in California had no facilities in the sample. The last set of rows in Table 1 shows that the production function for nursing homes is heavily labor intensive, with employee salaries and benefits accounting for 61 percent of total health care expenditures. Nurse salaries account for an average of about 61 percent of labor costs, with the remaining 35 percent spread across workers in plant operations and maintenance, housekeeping, dieticians, social coordinators, etc. The OSHPD data contains hours and salary information for 5 different nursing occupations: supervisors and managers; registered nurses (and geriatric nurse practitioners); licensed vocational nurses; nurse assistants (aides and orderlies); and other (including technicians, specialists, etc.). As shown in Figure 2, nursing assistants provided the lion s share of hours of direct care in 1999 at about 67 percent. Licensed vocational nurses accounted for about 18 percent, and registered nurses about 10 percent of hours of direct care. It should be noted that for licensed nurses, nursing homes constitute a small fraction of overall employment: in California nursing homes employed only about 3 percent of all RNs (hospitals employed 62 percent) 16 This is California s version of Medicaid. 13

16 and 21 percent of all LNs (hospitals employ 27 percent) in By contrast, nursing facilities employ about 41 percent of all nursing aides (hospitals employ 31 percent). 17 Finally, it should be noted that turnover of staff is a pervasive feature of nursing homes: half of all firms in the sample experienced average turnover rates over 64 percent. Figure 3 presents real hourly wages for each nursing occupation (excluding the other category) in 3 different years: 1996, 1999, and Several facts are evident in the Figure. First, there is a clear hierarchy across occupations with supervisors earning the most at an average of $31.78 per hour (excluding benefits) in 1999, registered nurses next at $23.84, followed by LVNs at $18.29 and nursing assistants at $9.50. These wage differentials reflect differences in preparation and training requirements for each occupation. RNs are typically required to complete between 2 and 6 years of post-secondary education, whereas LVNs typically require only one year. Nursing assistants require no formal training to be hired, but must complete 100 hours of on the job, and 50 hours of classroom training to be certified, and must pass a state medical exam within 4 months of being hired. Figure 2 also shows that real wages are increasing for all nurses through this time period, as the increase in the demand for nursing labor seems to have outpaced growth in supply. Finally, there is significant variation in average wages across firms for each occupation, though it is more pronounced for more skilled nurses. All other things equal, this type of variation in wages for the same type of worker has been used as prima facie evidence for monopsony in labor markets. Of course, all other things are not equal I return to this again below. 5 Empirical Results In this section I document the effects of the 2000 minimum staffing legislation for nursing homes in California. I first explore the effects on nurse staffing levels and composition at nursing homes, and then investigate whether these staffing effects were accompanied by wage effects that may be consistent with monopsony. In the final subsection I present 17 Data taken from California Employment Develepment Department LaborMarketInfo website: Accessed March 10,

17 estimates of the elasticity of labor supply for each type of nursing labor. In each section, results are also presented for an aggregate of hours worked by all non-nursing staff, as a false experiment to test for the influences of firm specific shocks that might bias the results. In each section I present non-parametric analyses of the raw data that illustrate the basic findings, and then refer to more parametric estimates of equations (1) and (2) presented above. 5.1 Effects of Legislation on Staffing in Nursing Homes Figure 4 shows the distribution of staffing levels across firms in selected years between 1996 and Each figure shows the number of facilities falling into hprd-wide bins of the staffing distribution relative to the 3.2 threshold shown with the vertical line. At the time the minimum staffing legislation was signed into law in 1999, only about one in four nursing homes had staffing levels high enough to be in compliance with the new 3.2 hprd standard that took effect in January of The distribution of staffing levels appears to have been quite steady since 1996 the first period in my data, but it began to rise almost immediately after the law took effect. As shown in the lower panel of Figure 4, two years after passage of the law in 2001 nearly half of all facilities were in compliance, and by 2003 more than 60 percent had hprd levels above the 3.2 threshold. While this break in the overall time series is suggestive of a causal effect of the law s passage on staffing levels, we can rule out the possibility that other forces drove up staffing levels by exploiting the fact that the law should have had a differentially large impact on firms in proportion to how far their initial staffing levels were below the standard. Figure 5 shows trends in the averages of total nurse hours employed (in log annual hours) for firms grouped according to deciles of their 1999 staffing levels in hprd. I group all firms with staffing ratios below the 3.2 hprd threshold in 1999 into 7 equal sized categories of 123 firms ranging from low hprd firms to those falling just short of the threshold. Similarly, I group all firms with staffing levels above the threshold into 3 groups of 90 firms each. I refer to these groupings of firms as deciles, though strictly speaking this is not accurate. Each 15

18 of the three panels in Figure 5 shows trends for the 5 even numbered (that is, the 2nd, 4th, 6th, 8th, and 10th) deciles of firms. The top panel shows trends in hours employed for nurse aides. We have already seen that nurse aides are by far the lowest cost category of nurse, and so it seems natural to expect firms to respond to a mandate to increase their total number of nursing hours by hiring more aides. The evidence in the figure is quite consistent with this prediction. Nearly all groups of firms showed little change or a mild decline in nurse aide hours before the minimum staffing legislation passed. Starting in 2000, one year after the legislation passed, however, all firms increased their employment of nurse aides. For firms that were already in compliance, those in the 8th and 10th deciles in the Figure, the increase in hours was quite mild on the order of 5-8 percent over four years. For the firms below the threshold, however, nurse aide employment increased significantly more, and the magnitude of the increase was largest up to 25 or 30 percent over four years for those firms furthest below the 3.2 hprd threshold (i.e., 2nd Decile firms). The fact that all groups of firms appear to follow similar trends in hours worked, coupled with the increase in hours worked proportionate to the gap between firms 1999 staffing level and the threshold lend support to the research design employed here. The lower two panels of Figure 5 reveal different patterns for more skilled nurses. For licensed vocational nurses (LVNs), the trends in employment are strikingly similar for all firms regardless of whether they were already in compliance with the new staffing standard. For each decile, there is a trend towards employing more LVNs throughout the period, with no noticeable acceleration after the passage of the staffing law. The lowest panel shows that RN employment is declining throughout the period in all groups of firms. After falling slightly up until 2000, it appears that this decline becomes more pronounced in most groups of firms. The subsequent decline seems highest, however, for firms with higher 1999 staffing levels and seems quite muted for firms with the lowest 1999 staffing levels (i.e., the 2nd and 4th Decile firms in the Figure). One plausible explanation is that pressures to reduce costs in the industry were driving substitution away from RNs towards less expensive LVNs and NAs. After the minimum staffing legislation, firms with the lowest staffing levels may 16

19 have been more constrained in being able to replace RNs, perhaps by Federal and other State regulations requiring minimum numbers of licensed and registered nurses to be on duty. From the standpoint of the research design above, it is again comforting to see fairly similar trends for all groups before the passage of the law since it makes it less plausible that unobserved factors influencing employment are trending in different ways for firms with different 1999 levels of staffing the key exclusion restriction for the instrumental variables approach to yield consistent inverse elasticity estimates. On the other hand, we can get a sense from Figure 5 that there appears to be a relationship between 1999 staffing levels and subsequent changes in employment for firms above the 3.2 threshold, especially for RNs. To control for the possibility that other factors may be creating changes in employment related to staffing levels, below I present estimates of the effect of the staffing law (captured by GAP ) controlling for a linear term in the facilities 1999 overall staffing level (in hprd). In any case, it appears that our instrumental variable strategy will work best for nurses aides, where the law had its clearest effect, but perhaps less well for RNs and much less well for LVNs. Figure 6 shows similar employment effects of the minimum staffing law for various nursing occupations in panels A through C, and for the aggregate of all non-nursing occupations in panel D, over a two year (1999 to 2001) and four year period (1999 to 2003). The dots in each picture represent each facility in the sample, with the change in log hours plotted on the y-axis and the facility s initial 1999 hprd staffing level on the x-axis. The solid line represents a locally weighted least squares fit through the data. Panel A of Figure 5 shows that the most dramatic effects of the law on staffing levels were apparent for nursing assistants. Facilities above the threshold saw little change in NA hours worked, though there is a mild negative relationship between the increase in nurse aide hours and staffing levels over the two year window, 1999 to The most striking aspect of the Figure, however, is a large increase in hours worked by NAs that is strongly related to how far below the staffing threshold a firm was before the law passed. For facilities in the lowest quintile of 1999 hprd staffing levels, nursing assistant hours increased by more than 20 percent. In 17

20 light of the fact that nursing assistants already comprise two-thirds or more of all hours of care at these facilities, this represents a significant increase in the overall number of hours of care. This visual evidence is borne out by regression evidence presented in the top half of Table 2. For each type of labor, different columns of the table present estimates of the effect of the minimum staffing law on changes in the log of total hours employed over one to four year periods following Estimates from three specifications are presented in the rows of the table. The row S1 refers to a simple bivariate regression of the change in log hours on the GAP i variable. Specification S2 adds fixed effects for each of 31 county groups, and S3 adds a control for a linear term in the firm s 1999 staffing level to S2. The first 4 columns of row S1 show that the law did in fact have a large impact on nurse aide employment. Since firms below the threshold had an average staffing level of about 2.72, the estimates imply effects for typical firms that were out of compliance ranging from an increase of 6.9 percent (( )*.138) in the first year to 12.2 percent over four years. While these estimates are not sensitive to the inclusion of county fixed effects, they are quite sensitive to the inclusion of a linear term in 1999 staffing levels. Since firms with higher staffing increase their staffing levels less even among firms above the threshold, controlling for this tends to reduce the estimated effect of the legislation. For nurses aides, the estimates for a typical firm range from about a five percent increase over one year to about a seven percent increase over four years. As in Figure 5, the effects of the law on licensed nurses is more ambiguous. In Panel B, we see an overall increase in LVN hours worked, but little pattern related to 1999 staffing levels. The estimates in the second four columns of Table 2 confirm that the law had no significant effect on their employment over any time window. Panel C shows how nursing facilities staffing of RNs changed due to the law. As the solid line is below the horizontal line at zero especially over the four year period 1999 to 2003, we see that on average, facilities were shedding RN staff. There is an overall downward slope to the line, suggesting that higher staffed facilities were shedding RN employment faster than firms with low staffing 18

21 levels. For the four year period, but not necessarily the two year period, it seems that this relationship is more pronounced (steeper) for firms that were below the 3.2 threshold when the law passed in Turning to the estimates in Table 2, then, we see that the estimates of the effect of the law on RN employment is quite sensitive to inclusion of the linear term in 1999 staffing. Without it (in specification S2), it appears that the law had the effect of preventing facilities with low initial RN staffing levels from reducing their RN staff further a firm below the threshold with average staffing had a change about 9 percent higher than firms right at the threshold. Controlling for 1999 staffing level reduces the magnitude of the effect to about 5 percent, and increases the standard error considerably. 18 Overall, the weight of the evidence suggests that the law caused nurse aide hours to increase considerably, and prevented a decline in RN hours among firms that were below the legislated minimum requirement. This has had obvious effects on the skill composition of the work force in nursing homes. In particular, as shown in Figure 7 the fraction of all hours worked by nursing assistants has risen (panel A). And due to the fact that LVN hours have not been affected by the law, the ratio of hours provided by licensed nurses (both RNs and LVNs) has declined. While the effects of staffing increases on patient outcomes will be left to a separate study, it is worth noting a tension lurking here. The goal of the minimum staffing policies is to improve patient care. While more overall hours would seem likely to further that goal, the degradation of the skill-level of people providing that care might mitigate some of the hoped-for benefit. Before moving on one might wonder whether there might be some other explanation for the pattern of hours changes we see here. In particular, are there other shocks to firm hiring that might be correlated in a similar way to 1999 hprd levels that might produce similar patterns? The evidence of panel D of Figure 6 suggests that such a story is unlikely. So long as these other shocks would likely affect all workers in nursing homes, we would expect them to affect non-nursing occupations in a similar fashion. As shown in the Figure, there s 18 Similar pictures for managers (not shown) reveal no change in their staffing levels, nor any relationship with initial staffing levels. 19

22 no change in the hours worked by non-nursing occupations, strongly suggesting that the patterns in the data discussed above can be causally attributed to the minimum staffing legislation. 5.2 Effects of Legislation on Nursing Wages in Nursing Homes Figure 8 shows analyses similar to those in Figure 6, depicting the effect of the minimum staffing legislation on nursing wages. The story here is quite straightforward, and we ve heard it already when discussing figure 3 above. For every occupational category, there are strong gains in wages overall ranging from 15 to 25 percent over 1999 to But the Figure reveals no apparent relationship between growth in wages and initial staffing levels. The bottom half of Table 2 amplifies this conclusion. The point estimates of the impact of the law on wages are small across all specifications, and for every type of labor. Even for nurse aides, where the largest increases in staffing were observed, the point estimates from specifications 2 and 3 suggest that an average firm with a 1999 staffing level of 2.7 increased their wages by between -.7 percent and 1.5 percent. None of the wage effect estimates are significantly different from zero in any specification, for any length of time, or for any class of worker. 5.3 Estimates of the Inverse Elasticity of Labor Supply to Nursing Homes The die has already been cast on the issue of monopsony in the market for nurses. The lack of any significant effects on wages suggests that we will not be able to reject the null that the inverse elasticity of labor supply for any group is zero. In discussing the instrumental variables results presented Table 3, I will thus focus on the range of parameters consistent with (the upper end of) the 95 percent confidence intervals. Before this, we should pause to note that the minimum staffing law appears only to have affected NA employment, and to a lesser extent RN employment. The policy appears to provide too weak an instrument for changes in hours worked by other occupations. The instrumental variable strategy employed here is therefore probably only informative about the markets for NAs and RNs, though 20