Chapter 5:  Factors Affecting the Supply of and Demand for LPNs

The labor market for licensed practical and vocational nurses consists of two components: the supply of LPNs and the demand for LPNs.  Both supply and demand should be affected by the wage paid to LPNs.  When wages rise, LPNs should find employment more attractive and increase their supply of labor.  Conversely, higher wages increase the cost of hiring to employers and thus demand should decline.  When there is a shortage or surplus of LPNs, wages should adjust to rectify the imbalance.

Numerous other factors can affect the supply of and demand for LPNs, however.  The family circumstances of LPNs may prohibit them from working full-time, and regulatory requirements might lead to higher or lower demand for LPNs.  This chapter examines the underlying supply of and demand for LPNs to identify the factors that affect LPNs’ decisions to work and employers’ demands for them.

The Supply of LPNs

A Conceptual Model of the LPN Supply

Labor markets for licensed nurses generally are not national in scope.  In some geographic regions there are few employers and these employers may have a high degree of control over the local labor market.  Other nursing labor markets are very competitive, with a plethora of employers.  Because job opportunities for licensed nurses are plentiful at nearly all times, nurses usually do not need to relocate to find interesting and rewarding work. 

The supply of nurses consists of nurses with active licenses.  Some of these nurses are not working in nursing, but they are part of the current pool of nurses potentially available to work.  The supply of nurses to a local labor market increases as nurses flow into the labor market by graduating from nursing programs, migrating from other regions, immigrating from other countries, or increasing hours worked.  The supply of nurses declines with retirements, migration out of the region, decreasing hours worked, and career changes out of nursing.  Figure 5.1 summarizes the labor flows in and out of the stock of licensed nurses.

The primary source of growth in the nursing workforce is graduations from nursing programs.  These graduations generally stem from interest in the nursing profession.  For the first part of the 20th century, licensed nursing was one of a few occupations widely open to women.  Most women faced limited career choices, and nursing was an attractive option to women who were interested in science.  As career opportunities expanded for women in the last quarter of the 20th century, however, nursing had to compete with numerous other attractive professions for new entrants.  It has been suggested that women now are less likely to choose a traditionally female-dominated career such as nursing (Buerhaus, Staiger, & Auerbach, 2000) .  However, an annual survey of 350,000 first-year college students across the U.S. found that the percent of students planning on a career in nursing remained steady at five percent between 1966 and 1996 (Astin, 1998).

Regional and international migration of LPNs has not been measured in any data sources of which we are aware.  The National Council of State Boards of Nursing does not maintain a national database of LPN licenses, and States do not link their licensure files so that LPNs can be tracked as they move from State to State.  LPNs do not exist in most other countries, so international migration of LPNs is not an important source of new LPNs.  This is reflected in the fact that relatively small and stable shares of LPNs are immigrants, as reported in Chapter 2.  Some registered nurses educated in other Nations do not pass the RN licensing board examination when they immigrate and subsequently take the LPN licensing examination.  To our knowledge, no source of data measures the extent to which this occurs. 

Figure 5.9: Flows and Stock of Licensed Practical/Vocational Nurses

The outflow from the supply of LPNs consists of nurses who retire, choose to permanently leave the profession, or who migrate to other countries or regions.  Unfortunately, there is no data with which one can examine any of these phenomena.  If a LPN allows his or her State license to lapse, it is not possible to identify whether the LPN obtained a license elsewhere, and thus we do not know if the LPN has left the supply of nurses.  LPNs who have active licenses but are not working are not identified in any national survey.  National data such as that collected by the Bureau of Labor Statistics and Bureau of the Census identify LPNs by their current occupation, and thus very few LPNs who are not working are identified in these data.

Thus, little can be said about important components of the inflow and outflow of LPNs.  The behavior of LPNs who are actively licensed and consider their current occupation to be that of LPN can be examined using the annual Current Population Survey conducted by the Bureau of Labor Statistics and the Bureau of the Census.  Many characteristics of these LPNs are available from these surveys, and the factors that affect labor supply can be considered in depth.

Data for Supply Analyses

We use data from the 1994-2001 Current Population Survey (CPS) Outgoing Rotation Group (ORG) (U.S. Bureau of the Census, 2004) to analyze factors that influence the supply of licensed practical nurses.  In order to identify licensed practical/vocational nurses in the Current Population Survey, we utilize the occupation codes.  With these codes, we identified 4,736 LPNs in the 1994-2001 CPS ORG files.  The resulting dataset used to estimate the supply of licensed practical nurses in the U.S. has 4,616 observations.  This number does not match the total number of LPNs in the 1994-2001 CPS ORG files since we delete LPN observations that have extreme values (defined as over the 99th percentile) for the earnings and work hours variables used in our analysis. 

Methods of Analysis

Economic theory suggests that an individual’s work decision is a function of individual (demographic) characteristics, family characteristics, and labor market conditions.  We use the Current Population Survey’s demographic and labor force information on LPNs to create variables for our models of the supply of LPNs.  The demographic variables in our models include the following: gender, age, educational attainment, race/ethnicity, and citizenship status.  Family characteristics included in our analysis are marital status, number of kids in household by age category (e.g. number of kids aged 0 to 5 in same household as LPN), and household earnings (defined as the sum of weekly earnings of all household members minus the LPN’s weekly earnings). 

Labor market variables were generated using the geographic and earnings data in the CPS.  We created dummy variables for each region in the United States (Northeast, Midwest, South, and West), and for the population size of the metropolitan statistical area in which LPNs in our sample reside.  Also included is the percentage of licensed practical nurses unionized in the LPN’s State of residence.  The market wage for LPNs is an important labor market condition.  We generate State-level market wages using hourly earnings from our sample of LPNs.  Because we had small numbers of observations for some States, we used a complex method to determine markets wages.  Each wage is based on 3 years of data, so the wage of a single year is the median of the wages of that year and the years immediately preceding and following that year.  For example, the market wage for 1990 is the median of the wages for 1989, 1990, and 1991. 

We then group LPN observations in each State based on whether they resided in a metropolitan statistical area (MSA).  Those residing in an MSA are considered to be living in an urban area, while those not residing in an MSA are considered to be in a rural area.  Using this information, we calculate urban and rural LPN wages for each State.  Since sample sizes were small for several States, we decided that the market wage associated with each LPN would have to be calculated from at least 15 observations.  We used the following algorithm to assign market wages: if LPN lives in an urban area in a State and the median urban wage for that State is calculated from at least 15 observations, then the market wage is the median urban wage; otherwise, the market wage is the State-level median wage.  Substituting “rural” for “urban” in the above algorithm explains the logic for assigning a market wage to LPNs residing in rural areas of a State.  Thus, we have three potential market wages for each State, but only one is matched to each LPN in our sample.

Even though we assume market wages are exogenous in our labor supply equations, we cannot rule out the possibility that they are determined simultaneously with supply, thus potentially biasing our estimates.  To address this concern, we use two-stage least squares regression as a specification check.  This technique produces predicted values for wages after estimating a wage equation. ^[2] We then use these predicted wages in our labor supply regressions, and compare the results with those from the regressions in which market wages are used.  As a third specification, we calculate wages for the LPNs in our sample who report being employed.  The CPS has data on usual weekly earnings and usual weekly hours of work.  We divide usual weekly earnings by usual weekly hours of work to obtain a measure of own wage for each LPN in our sample who reports being employed.  We then estimate the supply equations using own wages for working LPNs and predicted wages for non-working LPNs.  Thus, we run three regressions for each supply model, each with a different measure of wage.

We focused on three outcome measures in our analysis: (1) the probability of working (labor participation), (2) the probability of working full-time, defined as usually works 30 or more hours per week, and (3) usual hours of work per week.    We model each of these to examine the factors that affect the supply of licensed practical nurses.  Appendix E1 reports the means of the variables in the dataset used to estimate the supply of LPNs.  We discuss trends in the variables here.

Several of the demographic variables show an upward trend in their mean values during our sample time period. These variables include age, and the proportion of LPNs who are black, Native American, have completed some college, and hold an AA degree.  Those with a downward trend are the proportion of LPNs who are white and the percent that have no more than a high school education.  These trends were discussed in detail in Chapter 2.

The data show an increase in the percent of LPNs holding more than one job, usual hours worked per week, and usual weekly earnings before deductions.  Notably, the means of our wage variables follow a similar pattern over our sample time period.  They decrease until 1997 and then climb to near their 1994 values by 2001. Most of the market characteristics in the dataset exhibit a trend in their mean values.  Union representation/coverage of LPNs decreased, as did the share of LPNs residing in the Northeast and West, and the percent living in metropolitan areas with a population of 500,000 to 2.5 million.  The percent of LPNs in our sample that live in the South increased between 1994 and 2001, as did the proportion residing in rural areas. 

LPNs in our sample also increasingly worked for private employers, in personnel supply services, and the offices of physicians.  The share working for government and the percent who are self-employed declined during our sample time period.  The only family characteristic exhibiting a trend during our sample time period is household earnings, which increased between 1994 and 2001.

Factors That Affect the Employment of LPNs

Table 5.1 presents the estimated coefficients and marginal effects from probit regression equations of the likelihood of a LPN being employed using the Current Population Survey data for 1994 through 2001.  The marginal effect measures the increase in probability resulting from increases in the explanatory variable in the regression equation.  For example, the marginal effect of living in the Midwest is 0.016.  The explanatory variable has a value of 1 if an LPN lives in the Midwest and 0 otherwise.  Thus, living in the Midwest increases the probability of being employed 1.6 percentage points, which is the product of the marginal effect and the change in the explanatory variable.  In the regression equation tables, the statistical significance of the coefficients is indicated.  We focus our discussion on explanatory variables that are significant with a p-value of 0.05, meaning there is a 5 percent chance that the identified relationship is spurious.

The first three columns in Table 5.1 report the estimated coefficients, robust standard errors, and marginal effects for the regression in which market wages are included as an explanatory variable.  The next three columns report estimates for the two-stage least squares model in which predicted wages are used, and the final three columns report results from the regression in which the wage is defined separately, as described above, for working and non-working LPNs. From this point forward, we refer to this last measure of wage as “own wage.”

The results from the probit regression with market wages as an independent variable are quite similar to the results from the two-stage least squares regression in which predicted wages are used to estimate the supply model.  The probit regression in which own wages are used produce surprising results, especially concerning the effect of wage.

Though not statistically significant, the estimated coefficients on market wage and predicted wage and their squared values have the expected sign.  However, when estimating the model using own wages, we find a negative and statistically significant coefficient on wage.  The marginal effect implies that a one-dollar increase in wage decreases the likelihood of a LPN being employed by 1.4 percentage points.  Furthermore, the wage-squared coefficient is positive and statistically significant, implying that as the wage increases beyond a certain point, LPNs are more likely to work. This result is opposite the pattern found in many studies of labor supply.  The likelihood of employment typically rises with wage at nearly all wage levels.  It is important to note that the LPNs in our sample have very high labor participation rates, ranging from 92 percent to 96 percent during our sample time period of 1994-2001.  Thus, there is little variation in our outcome variable, and this may affect our regression results.  Nevertheless, several of the coefficients of the remaining explanatory variables across all three specifications of our model are in agreement with economic theory.

Demographic characteristics are important predictors of employment of LPNs.  The likelihood of working initially increases with age, by 0.1 to 0.4 percentage points, and then decreases as indicated by the coefficients on age squared.  The inflection points calculated from the marginal effects indicate that LPNs are less likely to work after age 38 (first specification), 40 (second specification), or 50 (third specification). Native American LPNs are 2.5 to 7.6 percentage points less likely to be working than white LPNs.  Black LPNs also are less likely to be employed, although the degree of statistical significance is lower in two of the specifications.  In contrast, Asian LPNs are more likely to be working, although this result is only statistically significant at a higher p-value.  LPNs who are US citizens by naturalization are 0.6 to 3.4 percentage points less likely to be employed than are US-born LPNs.  In the regression with market wage as an independent variable, LPNs who are not U.S. citizens also are less likely to be employed.

Family characteristics do not appear to be strong predictors of labor force participation.  In all three specifications of the model, only household earnings have a statistically significant relationship with the likelihood of working for LPNs.  LPNs are less likely to work as the earnings of other household members (such as the LPN’s spouse/partner) increase.  However, the marginal effects are practically zero. 

Table 5.1:  Probit Results for Probability of Working

*p < 0.10
**p < 0.05

Notes: (1) dependent variable equals one if employed, and equals zero otherwise; (2) all regressions include a constant; and (3) standard errors are estimated using the "robust" option in Stata.

Source: Current Population Survey Outgoing Rotation Group Files, 1994-2001

The labor market in which the LPN resides affects employment opportunities, and cultural differences across regions also may affect the likelihood of working.  As compared to LPNs living in the West, Midwest LPNs are 0.2 to 1.6 percentage points more likely to work. 

It is important to note that LPNs are identified by their self-reported occupation, and thus LPNs who are not working in nursing may not identify themselves as LPNs.  The CPS data thus likely overstate the probability of employment, and regression equations estimated for a broader sample of LPNs might produce different results

The Hours Worked by LPNs

Once an individual decides to work, a decision must be made about the extent to which to work.  Employees can work part-time or full-time, and the number of hours per week they work can vary.  Personal, family, and labor market characteristics affect the decision of how much to work.  To explore these relationships, we estimate regression equations similar to those estimated for whether a LPN is working. Table 5.2 presents probit regression equations for the probability of a LPN working full time (i.e., 30 or more hours per week).  Again we run three regressions, each with a different measure of wage.  The first specification, using market wages as an explanatory variable, is restricted to LPNs who report working, and thus the regression results only apply to the population of working LPNs.  The remaining specifications use the full sample of LPNs.  Despite differences in how we define the wage variable (and, thus, the wage-squared variable) in each of the three specifications of the model, the regression results are similar.

In all three specifications, the estimated coefficient on wage is positive.  It also is statistically significant except in the regression using predicted wages as an independent variable for all observations.  For the sample of working LPNs (specification (1)), a one-dollar increase in the market wage increases the likelihood of working full-time 6.8 percentage points.  In specification (3), a one-dollar increase in own wage increases the likelihood of full-time employment 2.6 percentage points.

Table 5.2:  Probit Results for Probability of Working Full-Time

*p < 0.10
**p < 0.05

Source: Current Population Survey Outgoing Rotation Group Files, 1994-2001

Notes: (1) dependent variable equals one if usually works 30+ hours per week, and equals zero otherwise; (2) in column one, sample is restricted to licensed practical/vocational nurses who reported being employed; (3) all regressions include a constant; and (4) standard errors are estimated using the "robust" option in Stata.

To check for the possibility of backward-bending supply, we included wage-squared as an independent variable.  The estimated coefficients are negative in all three specifications, and statistically significant in the regressions with market wages and own wages. The negative coefficients across the three specifications provide evidence that the labor supply of LPNs is backward bending, indicating that after a point, further wage increases reduce the likelihood of working full-time.  A possible explanation is that LPNs want to earn a target income, and as wages rise they need to work fewer hours to reach this target. 

Demographic characteristics are important predictors of whether LPNs work full-time.  Notably, the same demographic variables have statistically significant coefficients regardless of how we define wages.  Furthermore, there is very little difference in the marginal effects.  For example, black LPNs are 3.5 to 3.7 percentage points more likely to work full-time than are white LPNs.  Male LPNs are 7.0 to 7.2 percentage points more likely than females, and LPNs who are naturalized citizens are 8.3 to 8.6 percentage points more likely than U.S.-born LPNs.  LPNs with some college education but no degree are less likely to work full-time than LPNs who have never attended college.  Finally, LPNs are more likely to work full-time until their late thirties or early forties, after which time age has a negative association with the likelihood of working full-time.

Family characteristics also are important factors for LPNs in deciding whether to work full-time.  As the earnings of other members of the household increase, the likelihood of a LPN working full-time decreases.  However the estimated coefficients in all three specifications are small in magnitude and only the coefficient in the regression with market wages is statistically significant.  All three specifications of the model indicate that married LPNs are less likely to work full-time than are LPNs who have never been married.  As expected, the presence of children in the household is negatively associated with full-time work.  The results are similar for each age category and suggest that each child under the age of 18 reduces the likelihood of a LPN working full-time by approximately two percentage points. 

Several market characteristics affect the probability of a LPN working full-time.  LPNs residing in the South are 4.5 to 5.1 percentage points more likely to work full-time than are LPNs in the Western region of the U.S. The results for all three specifications of the model also indicate that LPNs residing in urban areas with a population between 100,000 and 499,999 are less likely to work full-time than those residing in less populated areas. Finally, compared to the beginning of the sample time period, LPNs in 2001 were more likely to work full-time.

Table 5.3 presents regression equations for the usual number of hours worked per week in the past year. As before, we run three regressions, each with a different measure of wage.  When market wages are used, the sample is restricted to LPNs who report being employed.  Otherwise, the full sample of working and non-working LPNs is used. 

The regression results are remarkably similar; however, there are key differences centered on the coefficients for wage.  In the specifications (1) and (2), wage is positively associated with hours of work.  However, this result is only statistically significant when we correct for the potential endogeneity of wages.  In this case, the estimated coefficient implies that LPNs on average work an additional 3.2 hours per week for each dollar increase in wage.  In specification (3), the coefficient on own wage is negative, but statistically insignificant.  Again, we find evidence of a backward bending supply curve.  In all three specifications, the estimated coefficient on wage squared is negative and statistically significant, albeit at a higher p-value. 

Male LPNs work more hours per week than do women, and black LPNs work more hours than white LPNs.  The number of hours worked increases with age until age 39 (37 in specification (3)) after which time age has a negative relationship with hours worked per week.  LPNs who are citizens by naturalization work an average of 2.5 to 2.6 hours per week more than do US-born LPNs.

Family characteristics affect the number of hours worked per week in ways that are consistent with the regression equations that examine full-time versus part-time work.  Married LPNs work approximately 2.2 fewer hours per week than do unmarried LPNs.  Children also reduce hours worked per week, with the effect being largest for children younger than thirteen.  The earnings of other members of a LPN’s household are negatively associated with hours worked per week, but in all specifications the size of the coefficient is so small as to be negligible. 

The average number of hours worked per week varies across regions of the United States.  Southern LPNs work 1.2 to 1.4 hours per week more than do LPNs in Western States, and LPNs living in the Northeast work fewer hours. 

The Demand for LPNs

The demand for licensed nurses is derived from the demand for health care, and is affected by a variety of factors, including the general population’s demographics and health, new medical treatments, health care payment systems, and health care regulations.  Health care providers rely on licensed nurses to provide the majority of direct patient care.  Registered nurses assess patients, develop plans for their care, perform tests, provide medical treatments, plan for patients’ discharges, teach patients and their families how to provide ongoing care, and keep detailed records of all these activities.  Licensed practical and vocational nurses assist in patient assessments and the development of care plans, provide medications to patients, start intravenous fluids, obtain blood samples, and participate in numerous other components of patient care.  Without licensed nurses, many health care providers could not care for patients. 

Table 5.3:  Regression Results for Usual Hours Worked Per Week

*p < 0.10
**p < 0.05

Source: Current Population Survey Outgoing Rotation Group Files, 1994-2001

Notes: (1) in the first column, the sample is restricted to nurses who reported being employed; (2) standard errors (in parentheses) are estimated using the “robust” option in Stata; and (3) all regressions include a constant. 

The dominant employer of licensed nurses is the hospital industry, although RNs are more likely to work in hospitals than are LPNs.  As the number of patients and patient days in hospitals rise, demand for RNs and LPNs rises (Spetz, 1999) .  The increasing acuity of illness of patients in the hospital makes RNs particularly important to hospital care, as does the diffusion of high-technology medical services in hospitals (Spetz, 1999) .  LPNs are generally restricted from giving patients medications through intravenous lines (IVs), administering blood products, and providing other types of care that are critical in the hospital setting.  These restrictions reduce the usefulness of LPNs to hospitals. 

A high share of LPNs work in nursing homes and long-term care facilities; relatively fewer RNs work in this setting.  Patients in nursing homes generally do not receive complex treatments such as IV medication therapy, and thus much of the patient care in nursing homes can be provided by LPNs and unlicensed nursing personnel.  LPNs assist in the ongoing assessment of nursing home patients and the administration of oral medications.  In this section we use hospital and nursing home data to examine the demand for LPNs by these employers.

Data for the Analysis of Hospital Demand

To analyze the demand for licensed practical/vocational nurses in general acute care hospitals, we use 1990-2000 data from the American Hospital Association (AHA) Annual Survey of Hospitals.  This database contains hospital-level information on organizational structure; facilities and services; community orientation; total beds, utilization, finances, and staffing; and location and other geographic codes.  The AHA surveys all hospitals in the United States and the response rate averages 85 to 95 percent annually (American Hospital Association, 1999) .  Thus, in any year, the AHA Annual Survey Database has around 6,000 hospital observations.

The AHA Annual Survey asks hospitals to report full-time and part-time personnel for the total facility and for specific types of personnel, including registered nurses and licensed practical/vocational nurses.  The survey specifically defines full-time as working 35 hours or more per week, and part-time as working less than 35 hours per week (American Hospital Association, 1999) .  The staffing figures reported by the hospitals are then converted by the AHA into full-time equivalent (FTE) measures.  According to the AHA, full-time equivalent figures are calculated by adding the number of full-time personnel to half the number of part-time personnel (American Hospital Association, 1999) .  We use full-time equivalent LPN employment as our measure of LPN staffing for short-term, general acute care hospitals.  However, we should note that this measure potentially overestimates or underestimates the use of LPNs by hospitals.  For example, a nurse who works 20 hours per week and one who works 34 hours per week each would be counted as one-half of an FTE.  Similarly, a nurse who works 35 hours per week and one who works 40 hours per week would each count as one FTE. 

We model hospital demand for LPNs as a function of hospital, patient, and market characteristics.  This model is similar to that used in previous studies of the demand for nurses (Spetz, 1999) .  We construct hospital characteristic variables using data from the AHA.  We measure the quantity of care provided by each hospital in our sample as the number of patient days.  Also included in our model are Medicare’s share of total patient days, and the hospital’s service mix.  Our measure of service mix is the Saidin technology index (Spetz and Maiuro, 2004) .  The Saidin index provides a measure of the degree of technology available at hospitals by weighting each potential service and calculating the sum of weighted services available at each hospital.  The more rare the technology used by a hospital, the higher the weight it receives (Spetz & Maiuro, 2004) .

Patient characteristics in our demand model are the average length of stay (available from the AHA data) and the hospital’s case mix index from Medicare files (available from the Center for Medicare & Medicaid Services).  Both measures control for changes in patient volume, but the case mix index also controls for variation in the complexity or severity of cases treated by hospitals. 

We use data from the 1989-2001 Current Population Survey Outgoing Rotation files and the Bureau of Health Professions Area Resource File (ARF) (Bureau of the Health Professions, 2003) to create market-level variables.  The CPS contains union status information and we use this to create variables denoting the percentage of LPNs, RNs, and all workers in a given State who are covered by or a member of a union.  We calculate market wages for registered nurses, licensed practical nurses, and nurse aides using earnings data from the CPS ORG files.  The market wages are median values calculated from 3 years of data.  For example, 1990 LPN market wages are based on hourly earnings reported by LPNs in 1989, 1990, and 1991.  Furthermore, we calculate these at both the State level and for urban and rural areas within a State.  Thus, for each nurse type, we have with three potential market wages per State.  We attach an LPN, RN, and nurse aide market wage to each hospital observation in our sample depending on the number of observations used in creating the respective market wage.  If the rural or urban wage for a given State was calculated from less than 15 observations, then we assign the State-level wage to the hospital.  Otherwise, we assign the rural wage if the hospital is in a rural area or the urban wage if the hospital is in an urban area.  In the end, each hospital observation in our sample is matched to three market wages, one for each type of nurse.

We also include managed care variables in our demand model, which were generously provided by Douglas R. Wholey of the University of Minnesota.  Managed care activity is measured with two variables: the number of HMOs operating in the county and HMO penetration.  We also create a variable interacting these two measures of the managed care environment, and include this in our analysis (Wholey, Christianson, Engberg, & Bryce, 1997) .  County-level per capita income also is included in the model, and was obtained from the Area Resource File.  Finally, we include the two State-level scope of practice variables described in Chapter 3 in some equations.

We estimated our demand equations including several other variables from the ARF, such as physicians per 1,000 population and the share of population estimated to be aged 65 and over; however, we do not report the results of these regressions because these variables had no statistically significant relationship with our dependent variable, nor did their inclusion affect any other coefficients.  Our dataset for estimating hospital demand for licensed practical nurses contains 54,258 hospital observations over our sample time period from 1990 to 2000. 

As shown in Appendix E2, the average number of full-time equivalent LPNs in our sample of hospitals declined between 1990 and 2000.  In contrast, the mean number of full-time equivalent RNs increased.  As a result of these trends, the ratio of LPNs to all licensed nurses declined during our sample time period. 

All of the variables denoting hospital and patient characteristics exhibit trends in their mean values.  The average number of inpatient days and length of stay declined between 1990 and 2000.  Medicaid’s share of inpatient days increased, however, as did the service mix and the severity of cases treated in our sample of hospitals. 

Market wages for LPNs, RNs, and nurse aides were higher on average in 2000 compared to 1990.  However, the data do not show a continuous upward trend during our sample time period.  RN and LPN market wages increased between 1990 and 1994, and then declined during the mid-1990s.  In contrast, market wages for nurse aides declined during the first half of our sample time period, and then increased between 1994 and 2000.

Other market characteristics in our dataset also exhibit trends.  The degree of HMO penetration increased between 1990 and 2000, as did the average number of HMOs operating in a county.  In addition, the average per capita income in the hospitals’ counties increased during our sample time period.

Methods for Analyzing Hospital Demand for LPNs

In our hospital demand analysis, our dependent variable is the log of the number of full-time equivalent LPNs.  We also log several of our independent variables to normalize their distributions.  Thus, our demand equation is log-linear in form.  Each regression includes dummy variables for each year in our sample.  We estimate robust standard errors using the “cluster” command in Stata because it is possible that observations within a State may not be independent (StataCorp, 2003) .

We use several estimation methods in our demand analysis.  This is motivated by two concerns.  One is that there could be some unknown factor inherent to each hospital that affects its demand for licensed practical nurses.  If this is the case, ordinary least squares (OLS) estimates will be inefficient.  To address this possibility, we estimate fixed effects models to allow for individual hospital effects. 

Another concern is the potential endogeniety of LPN wages¹.  If wages are endogenous in the demand equation, then OLS estimates will be inconsistent.  Thus, we also estimate our demand equation using the instrumental variable procedure in Stata (StataCorp, 2003).  To use this procedure, we have to find variables that are correlated with wages, but not correlated with the error term in our demand equation.  County unemployment rates, obtained from the Area Resource File, have been used as an instrument for nurse wages in other studies (Spetz, 1999) .  As unemployment rates rise, spouses are more likely to be unemployed, and thus the nurse is more likely to work.  We also try two other instruments: the average age of LPNs in the hospital’s market area², and the percent of all workers unionized within the State.  We estimate first-stage regressions for LPN wages including these instruments as explanatory variables, and consistently find that the estimated coefficients on all but the county-level unemployment rates are highly significant.  Thus, we determine that the average age of LPNs and the percent of workers unionized within a State are good instruments for LPN wage in our demand equation.  We further check for the endogeneity of wages by conducting a Hausman test (Hausman, 1978; Kennedy, 1998; StataCorp, 2003) .  The test results provide no evidence that LPN market wages are endogenous in our model.  Thus, we report regression results both with and without instrumental variables, because although theory suggests instrumental variables are needed, the Hausman test indicates they may not be appropriate.

Longitudinal Analysis of Hospital Demand for LPNs

Table 5.4 presents regression equations estimating hospital demand for licensed practical nurses as a function of hospital, patient, and market characteristics.  The first two columns present the ordinary least squares equation coefficients and standard errors.  The second two columns present the results of a fixed effects equation, which includes a dummy variable for each hospital to control for hospital characteristics that are unobserved and constant over time.  The final two columns contain the results of the model estimated with fixed effects and instruments to control for the endogeneity of wages.

Conventional economic theory predicts that demand for employees will decline as their wages rise.  At the same time, demand for a type of employee could rise or fall with the wages of other employees, depending on whether other employees are complements or substitutes.  The results presented in Table 5.4 are consistent with this theory.  Higher LPN wages have a negative effect on demand for LPNs when instrumental variables are used to control for the endogeneity of wages.  The importance of addressing endogeneity is demonstrated by the positive, significant relationship between wages and demand in the uninstrumented fixed effects model.  In all three models, higher RN wages are associated with higher demand for LPNs. This finding suggests that LPNs are used as substitutes for RNs, at least in part.  The fixed effects and instrumental variables models indicate that a ten percent increase in the RN wage will raise LPN demand about two to three percent.  Aide wages have a modest positive relationship to demand for LPNs in the fixed effects equations, with a ten percent increase in the aide wage having less than a one percent effect on demand.  In the ordinary least squares equation, the aide wage has a very large, negative effect on LPN demand.

The volume of patients cared for at a hospital has an important effect on demand for LPNs.  The fixed effects and instrumental variables models estimate that ten percent growth in the number of inpatient days increases the demand for LPNs by about four percent.  Conversely, as the length of stay of these patients rises, the demand for LPNs falls.  The coefficient measuring the relationship between case mix and demand for LPNs is negative as well.  LPNs are less able to care for acutely ill patients, and thus as acuity rises, demand will fall.  Hospitals with a higher level of technology demand fewer LPNs. 

The ability of hospitals to hire staff depends on the revenue received in exchange for patient care services.  Several variables measure the potential revenues available to hospitals.  As the share of patient days reimbursed by Medicaid rises, demand for LPNs also rises.  Medicaid reimbursements to hospitals are known to be low, and hospitals that have high shares of Medicaid patients also typically have large shares of charity and non-paying patients.  Thus, it is possible that this relationship results from hospitals with a high share of Medicaid patients having smaller personnel budgets.  Another possibility is that Medicaid patients are somewhat less acutely ill than are other patients, and thus as the share of Medicaid patients rises, LPNs are better able to care for more patients.

The next three variables measure the relationship between the type of hospital owner and demand for LPNs.  For-profit, district, and government hospitals have greater demand for LPNs than do not-for-profit hospitals, holding other factors constant.  The potential reasons for these findings vary by type of owner.  For-profit hospitals have a financial incentive to hire less-expensive LPNs to increase their profit margins.  District and government hospitals may have smaller personnel budgets because they rely at least in part on tax revenues; thus, they may stretch their budgets with LPNs.

Table 5.4:  Estimates of Demand for Licensed Practical/Vocational Nurses in U.S. General Acute Care Hospitals, 1990-2000

*p < 0.10
**p < 0.05

Sources: American Hospital Association Annual Survey of Hospitals, Current Population Survey Outgoing Rotation Group Files, and Area Resource File.  Managed care data courtesy of Douglas R. Wholey

Notes: (1) the dependent variable is log (Number of Full-time Equivalent Licensed Practical Nurses) (2) all regressions include a constant; and (3) OLS regression uses the cluster (on State) option in Stata.

As HMO penetration and the number of HMOs operating in a county rise, the demand for LPNs falls, and these effects are somewhat accelerated as the interaction between penetration and the number of HMOs rises.  Greater HMO penetration in a market is thought to have a primary effect of reducing revenues available to hospitals.  Such revenue reduction could reduce demand for LPNs because hospital budgets are tighter.  However, HMOs also may value the quality of care offered by hospitals, and thus as HMO penetration increases, hospitals are pressured to favor the hiring of more-skilled RNs while reducing LPN staff.

County income affects demand for LPNs.  As per capita income rises, the demand for LPNs falls.  This relationship may arise if wealthier patients prefer hospitals with more highly skilled staff, and thus hospital demand for LPNs falls.

Statewide unionization of LPNs is associated with greater demand for LPNs in the instrumental variables equation.  This relationship may indicate that unionized LPNs are better able to ensure that they are retained in hospital staffing models. Conversely, LPNs may be more likely to unionize when their numbers are higher in the hospital industry.  RN unionization has no statistically significant relationship to LPN demand.

The coefficients of the yearly dummy variables indicate that there has been some change in hospital demand for LPNs over time. In 1993, demand for LPNs was lower than in 1990, while demand rose from 1995 through 1999.  This period of increased demand coincides with reports that hospitals were redesigning their nursing services to emphasize team nursing and less-skilled nursing personnel.  In these staffing strategies, LPNs would have had a more prominent role, and thus demand for LPNs would have risen. 

Table 5.5 presents regression equations similar to Table 5.4, but the dependent variable is employment of LPNs as a share of all licensed nurses.  In these equations, we can directly compare the effects of explanatory variables on demand for LPNs to demand for RNs.  The results confirm those of the level of LPN employment equations.  Relative demand for LPNs declines as the LPN wage rises, and it rises with growth in RN wages. 

Increases in the number of inpatient days has no effect on relative demand for LPNs, suggesting that hospitals maintain a consistent skill mix even as patient volumes rise.  Longer lengths of patient stays increase relative demand for LPNs, even though they decrease overall demand for LPNs.  Together, these findings suggest that longer lengths of stay are associated with lower overall demand for nursing care, perhaps because the share of patients in intermediate and rehabilitation units increases.

A higher patient case mix reduces relative demand for LPNs, although this relationship is statistically significant only in the ordinary least squares equation.  The coefficient on the technology index is consistent with expectations, in that higher technology reduces relative demand for LPNs.  It is possible that case mix is collinear with both length of stay and the technology index, so the statistically insignificant coefficients for case mix result from multicollinearity rather than a lack of relationship.

Table 5.5:  Estimates of Relative Demand for Licensed Practical/Vocational Nurses

*p < 0.10
**p < 0.05

Notes: (1) the dependent variable is log (LPNs as a Proportion of Total Licensed Nurse Staff) (2) all regressions include a constant; and (3) OLS regression uses the cluster (on State) option in Stata.

Sources: American Hospital Association Annual Survey of Hospitals, Current Population Survey Outgoing Rotation Group Files, and Area Resource File.  Managed care data courtesy of Douglas R. Wholey.

The effects of payer mix and hospital ownership in the relative demand equations are similar to those in the level of demand equations.  Hospitals with higher shares of Medicaid inpatient days have greater relative demand for LPNs, and the relative demand for LPNs falls as HMO penetration and the number of HMOs increases.  For-profit, district, and government hospitals have greater demand for LPNs relative to RNs than not-for-profit hospitals.  Per capita county income also has a negative effect on relative demand for LPNs.  Hospitals in States with higher shares of LPNs in unions have greater relative demand for LPNs.

Relative demand for LPNs declined from 1991 through 2000 (relative to 1990).  Combined with Table 5.4, these findings indicate that although absolute demand for LPNs stabilized in the late 1990s, hospitals have demanded relatively more RNs over time.

These findings demonstrate the importance of wages, hospital characteristics, and payer mix on hospital demand for LPNs.  As hospitals face increased pressure to reduce costs, or face higher wages for RNs and LPNs, the demand for LPNs changes significantly.  There have been periods of time during which LPNs have been considered attractive substitutes for RNs, and other times when demand for LPNs dropped because hospitals preferred RNs.  These demand changes have large effects on the career opportunities of LPNs. 

The Effect of Scope of Practice on Hospital Demand for LPNs

The longitudinal models presented above omit one important factor that could affect demand for LPNs: scope of practice regulations. Using the categorizations of LPN scope of practice created as part of this study, we examined the relationship between the scope of practice of LPNs and hospital demand for LPNs.  This is a complex undertaking, because these things are determined jointly.  For example, a liberal scope of practice may encourage employers to demand LPNs and reduce demand for other workers such as RNs.  However, when there is a shortage of RNs, employers are likely to increase their demand for LPNs and also to lobby State legislatures for expanded scope of practice for LPNs.  Because the relationship between demand and scope of practice is likely to be endogenous, we use instrumental variables to predict scope of practice regulations, in a fashion similar to that used to control for endogeneity of wages.  Our instruments are a set of variables measuring the political characteristics of each State: whether there is Democratic control of both legislative houses and the governorship, whether there is divided control of the legislature and/or governorship, the ratio of per capita State debt to per capita income, whether the governor has a line item veto, the percent of the upper legislative house that is Democratic, and the percent of the lower legislative house that is Democratic.  Mark W. Smith from the Veterans Health Administration Health Economics Resource Center in Menlo Park kindly provided these variables.

Because we have scope of practice data for only 1 year, we estimate the demand for LPNs using only a single year of data.  Table 5.6 presents the results of regression equations for hospital demand for LPNs using data from 2000, and Table 5.7 presents analogous equations for relative demand for LPNs (as a share of total licensed nurse employment).  The tables are organized in the same way as Tables 5.4 and 5.5.  As seen in the first two rows of Table 5.6, hospitals in States with restrictive scopes of LPN practice tend to have lower employment of LPNs.  However, once the potential endogeneity of wages and scope of practice are addressed using instrumental variables, the relationship is no longer statistically significant.  A similar pattern holds for the specificity of scope of practice.  However, Table 5.7 demonstrates that as the scope of practice of LPNs becomes more restrictive, the demand for LPNs falls relative to the demand for all licensed nurses, even when controlling for the endogeneity of scope of practice.

There are some differences in the effects of other explanatory variables between the cross-section and longitudinal results.  LPN wages continue to have a negative effect on demand for LPNs, but this effect is not significant when instrumental variables are used to control for the endogeneity of LPN wages.  RN and aide wages are not significantly associated with LPN demand, except in the uninstrumented equations.  In these equations, higher aide wages are associated with greater demand for LPNs.  As seen in Table 5.7, wages have little to no effect on relative demand for LPNs.

Table 5.6:  Estimates of Demand for Licensed Practical/Vocational Nurses in U.S. General Acute Care Hospitals, 2000

*p < 0.10
**p < 0.05

Notes: (1) dependent variable is log (No. of Full-time Equivalent Licensed Practical Nurses), (2) all regressions include State dummy variables and a constant; and (3) all regressions use the cluster (on State) option in Stata.

Sources: American Hospital Association Annual Survey of Hospitals, Current Population Survey Outgoing Rotation Group Files, and Area Resource File.  Managed care data courtesy of Douglas R. Wholey Political variables courtesy of Mark W. Smith, Health Economics Resource Center, VA Palo Alto Health Care System.

Higher patient volumes increase the demand for LPNs, and this relationship is larger in magnitude in the cross-section than it was in the longitudinal data.  However, higher volumes reduce the relative demand for LPNs in the cross section, suggesting that larger hospitals demand fewer LPNs, all other things held equal.  LPN demand is negatively associated with length of stay, but relative demand for LPN rises with length of stay, again suggesting that the acuity of patients declines with length of stay.  Thus, both overall demand for nursing staff and demand for RNs drops as length of stay rises.  Relative demand for LPNs falls as the case mix of patients rises.

Table 5.7:  Estimates of Demand for Licensed Practical/Vocational Nurses in U.S. General Acute Care Hospitals, 2000