Abstract
The purpose of this study was to investigate the potential relationship between student enrollment trends in elective secondary music ensembles and music ensemble teacher job turnover. Although student enrollment is widely accepted as an important concern for music educators and a crude proxy measure of music teacher quality, these normative beliefs have not been thoroughly examined empirically. This study tested these beliefs using data from a State Longitudinal Data System to link statewide high school student ensemble enrollment data to teacher workforce data for the academic years 2012 to 2013 through 2019 to 2020. Two-way fixed effects estimators with logistic and multinomial logistic regression showed that decreasing enrollments in high school music ensembles predict music teachers’ departure from the profession. A comparative interrupted time-series analysis showed that a change in music teacher does not significantly affect the future enrollment trend of a high school music ensemble program. An exploratory analysis examining the postteaching careers of former high school music teachers showed that the majority of music teachers who exited the profession earned considerably higher wages in their new careers. The authors conclude by discussing the implications of the results for music teachers, music administrators, music teacher education, and future research.
American high school music educators are often preternaturally concerned with their “numbers” and how their “numbers” fluctuate from year to year. This concern is often so acutely felt that an insider audience of music educators reading this paragraph would likely not need to be told that by “numbers,” we are referring to the annual student enrollment count that is often made available to the music educator shortly after the student course selection process closes in the spring of each school year. Recruitment and retention of students in the secondary music program are seen as so fundamental to the work of a secondary music teacher that these topics are often discussed in preservice music education textbooks (e.g., Brinson & Demorest, 2012; Colwell et al., 2017), practitioner journals (e.g., Cole, 2010; Marra, 2022; Pendergast, 2021; Sussman, 2010), and the research literature (e.g., Corenblum & Marshall, 1998; Kinney, 2019; Pendergast, 2020; Rinn, 2023). The professional obsession with student enrollment from year to year is not without just cause. Often, music teachers’ employment or full-time status is directly linked to the number of students who elect to take their courses.
Absent a professional consensus on metrics that link music teacher effectiveness to students’ ability to demonstrate learning of musical skills, grasp musical concepts, and engage aesthetically with the music they rehearse and perform, the size of a secondary music program in terms of student enrollment has come to serve as a substitute indicator for the quality of the music instruction offered within the program. Essentially, this means that the proportion of students attending a given school who are enrolled in the music program is often perceived by district and building administrators, other music educators, and community members as a reasonable proxy for the effectiveness of the music teacher. Fitzpatrick (2011) formalized this proxy into the research literature by categorizing music programs as “thriving” or “struggling” based on the program’s enrollment numbers, the reputation of the program within its community, and the program’s participation in music contests and festivals. Likewise, Major (2013) showed that district and school administrators consider student enrollment numbers when making consequential and existential policy decisions regarding elective music programs, with one administrator participant simply stating to her that “students vote with their feet.”
Although the emphasis on student enrollment as a key concern for music educators in the United States and a proxy measure for American music teacher quality seems to occupy a revered place within the profession’s lore, these normative beliefs have not yet been subject to strict empirical scrutiny. Building on professional lore to develop an empirically testable theoretical proposition requires us to consider the broader implications for student enrollment’s possible link to music teacher quality or effectiveness and how this link might play out in empirically observable ways. This thought experiment requires us to ask ourselves, “How might a thriving program—that is, one with relatively high or increasing enrollment—manifest its success in ways beyond enrollment that are observable?” One possible way that the success of a thriving music program might manifest beyond enrollment is through the retention of the teacher leading it. If a music program is successful, it is reasonable to assume that the music teacher may be more likely to remain in that position. Leading a thriving music program could lead music teachers to feel a greater sense of satisfaction and fulfillment from their work, and the recognition of the success might yield desirable career outcomes such as increased funding for the program, greater community support, or professional accolades for the teacher. Conversely, struggling music programs—those with low or decreasing enrollment—may experience greater teacher turnover rates because teachers leading them may feel frustrated or unsupported in their efforts to teach effectively. Reduced student enrollment might lead to fewer resources for the program, which might, in turn, create an increasingly difficult working environment leading to a teacher’s resignation or removal.
Given that teacher turnover can significantly disrupt student learning and, indeed, the operation of a school, it is unsurprising that a fair amount of research attention has been paid to the issue of teacher turnover and attrition in education broadly and music education specifically. The teacher turnover literature uses a tripartite classification to describe the outcome of a teacher’s employment continuation from one academic year to the next: Those who remain in the same teaching assignment at the same school are considered “stayers,” those who remain in the teaching profession but do so at a different school or district or in a different subject area are considered “movers,” and those who end their teaching careers entirely and transition to other work or retirement are considered “leavers.” Teachers who are part of a “turnover,” then, are those that would be classified as movers or leavers because changing from specific teaching positions at the end of an academic year classically defines teacher turnover. This categorization for teacher turnover outcomes appears to have its origins in the earliest incarnations of the National Center for Education Statistics Schools and Staffing Survey/Teacher Followup Survey program (Bobbitt et al., 1991) and has in the intervening 40 years become widely used both within music education and beyond music education (e.g., Kim, 2019; Nguyen, 2021).
The research on teacher turnover suggests several interrelated personal and school factors are related to teachers moving between jobs or leaving the teaching profession (Boe et al., 1997). Among the most noteworthy factors are workplace satisfaction (Borman & Dowling, 2008; Boyd et al., 2010; Loeb et al., 2005; Nguyen & Springer, 2023; Nguyen et al., 2019), perceived administrative support (Kim, 2019; Nguyen, 2021; Nguyen et al., 2019; Nguyen & Springer, 2023; Weiss, 1999), and teacher effectiveness (Boyd et al., 2010; Nguyen & Springer, 2023; Vagi et al., 2019). It is perhaps unsurprising that teachers who reported higher levels of workplace satisfaction and administrator support were less likely to leave their jobs than their less satisfied colleagues. Indeed, these factors have consistently been the subject of study within the music teacher labor market literature, and music education researchers have consistently affirmed these relations (e.g., Gardner, 2010; Hancock, 2008, 2009; Robison & Russell, 2020; Russell, 2012). Furthermore, one common experience reported by successful music teachers is the desire to move positions from one seen as less prestigious to one seen as more prestigious as a form of career advancement or what Hancock (2016) termed “promotion through relocation” (p. 432). However, when compared to the findings in the broader teacher literature, the music teacher turnover literature diverges in its relative lack of attention to teacher effectiveness as a predicate of turnover. In general education research, teacher effectiveness is typically measured by value-added measures and student achievement on standardized tests. The validity of these measures as indicators of teacher effectiveness in areas outside of music is still somewhat contested, but it remains that in the music education literature, there has emerged no clear consensus on how to authentically measure student learning in music, and absent accepted measures of student learning, measuring music teacher effectiveness remains elusive.
Taken together, the current state of the research literature regarding teacher turnover within and beyond music education undoubtedly suggests a clear gap in the research within music. Although music education research has explored two of the three key themes (workplace satisfaction and administrator support), the lack of teacher effectiveness measures has hampered researchers from linking music teacher effectiveness to job turnover. Absent an accepted measure of secondary ensemble music teacher effectiveness, we are left with the profession’s insider normative beliefs that elective student enrollment counts are a widely accepted (if somewhat colloquial) shorthand proxy for the effectiveness of a music teacher. Empirical work linking elective student enrollment trends to music teacher turnover could serve the state of the research literature in two key ways, depending on how one interprets the logic underpinning a theoretical link. For example, we might take it as a premise that music teachers’ normative beliefs are correct (higher enrollment truly means a more effective teacher). If this is the case, then demonstrating an empirical relation between declining student enrollment and music teacher attrition replicates, in music, the research findings from the broader education literature showing that more effective teachers are less likely to leave their jobs. Alternatively, we might instead take as a premise that the research suggesting more effective teachers are less likely to leave their jobs is applicable within music (and there is little in the research literature to suggest otherwise). In that case, demonstrating a relation between declining student enrollment and music teachers leaving the profession or moving jobs would provide an empirical undergirding to the profession’s normative beliefs.
One reason this line of thinking has not been explored has been, to date, the absence of easily accessible data sources that would allow for a reasonable investigation. The design of such a study is relatively easy to devise: track student enrollment for a number of secondary music teachers leading elective programs across a defined period of time and observe whether the teachers stayed, left, or moved jobs at the end of the time period. What is a relatively easy design to propose can become an insurmountable challenge in practice: Administrative data tracking accurate enrollment counts for even one school’s music program—linked, no less, to the specific teacher directing the program—would be difficult for a music education researcher to obtain, and the sheer number of different schools one would have to include in the study would exponentially increase the difficulty.
Research questions of this type, where the “thought experiment” of the design is straightforward but the aggregate data access would be prohibitive, were just one of many reasons that Congress, in the same statute that established the U.S. Department of Education’s Institute of Education Sciences (IES), established a grant program to incentivize states to develop statewide longitudinal data systems (SLDSs) to warehouse data on students and workers from many different sources. Initially, the SLDS program began as a response to political calls for greater transparency and accountability in education through data-driven decision-making, or as IES (2023) sloganized it: “Better decisions require better information.” The SLDS initiative aimed to support states in developing longitudinal data systems that track individual student progress and outcomes across multiple years and across education levels, from early childhood through postsecondary education and into the workforce. Since the initial disbursements of 14 awards in 2006, the SLDS program has undergone multiple iterations and updates, expanding its scope and providing states with technical assistance through the SLDS State Support Team to improve data quality, privacy, and security.
As it happens, an SLDS is the ideal source of exactly the kind of multisite, multiteacher, census-style administrative data to empirically explore the question of how student elective enrollment might be related to music teacher turnover. Through the information stored in a longitudinal data system, we can observe individual students as they elect various kinds of music ensembles throughout their secondary school years and observe which specific teachers were teaching the elective music courses in which the students enrolled. Thus, using educational data in the SLDS, we can calculate the precise sum total of students enrolled in each teacher’s ensemble program for a particular year and repeat this calculation for every music ensemble teacher working in the entire state. Given the longitudinal nature of the data, we can repeat these calculations annually for a period of time. If the SLDS also warehouses workforce information, we can then link the educational enrollment data to the teachers’ job dispositions—in other words, using SLDS data, we can observe whether any particular teacher was a mover, a leaver, or a stayer. The availability of these data, therefore, allows us to apply empirical scrutiny to the long-standing professional conventional wisdom that successful music teachers are successful at increasing or maintaining student elective music enrollment.
Purpose of the Study and Research Questions
The purpose of this study was to investigate the potential relationship between student enrollment trends in elective secondary music ensembles and music ensemble teacher job turnover. Given the lack of prior empirical literature on this topic, we examine the relation in both causal directions—whether declining enrollment predicts teacher turnover and whether teacher turnover affects future enrollment. The primary research questions that guided this inquiry were: (1) To what extent do student enrollment trends predict music teacher turnover outcomes? and (2) To what extent does high school music teacher turnover impact music program enrollment in years subsequent to the teacher change? Additionally, we pursued the following question in an exploratory analysis: (3) What careers do high school music teachers pursue after they leave the teacher workforce?
Method
Data Source
We leveraged data from the Maryland Longitudinal Data System (MLDS) Center (https://mldscenter.maryland.gov) spanning academic years 2012–2013 to 2019–2020 to answer our research questions. Like many SLDSs, the MLDS is a data warehouse designed to track the educational and workforce progress of all students enrolled in Maryland’s public schools from kindergarten through 12th grade into postsecondary education at schools located in Maryland or in any of the other U.S. states and into the workforce. Although complete longitudinal data are available only for students who begin schooling in Maryland and return or remain in the state for their adult employment, cross-sectionally, the K–12 and workforce data are essentially censuses: All Maryland students and all Maryland workers in a given cohort year are observable in the data. Crucially, the cross-sector nature of the data warehoused in the MLDS allows us not only to observe individual students from the entire state of Maryland who were enrolled in high school music programs but also to link those students to specific music teachers employed to teach them. We linked student course enrollment data from the Maryland State Department of Education (MSDE) with teacher data and wage data from the Maryland Department of Labor to conduct our cross-sector analysis. Importantly, given the sheer amount of personally identifiable information located within the MLDS Data, use of the data is restricted and subject to both preapproval of the research questions and a strict nondisclosure agreement. These protections are in place to prevent the inadvertent release of student or adult personally identifiable information; both our analytic plan and the results arising from our analyses underwent a required disclosure risk review prior to the publication of this article.
Our first tasks were to identify high school music ensemble courses taken by students in the data set, link those students to their music ensemble teacher, and devise an annual enrollment count for each teacher’s program. Following Miller (2023b), we identified high school ensemble music courses that appeared on student transcripts both through the School Courses for the Exchange of Data (SCED) codes and local education agency (LEA) verbatim course names. Then, we computed student enrollment numbers by summing the number of unique students enrolled in each program by year. Because of LEA variation in course naming conventions and SCED code usage, we compiled total aggregate program enrollment data rather than enrollment by individual courses. For example, if a school offered two wind band courses taught by the same teacher named “Concert Band” and “Symphonic Band,” we combined the enrollment counts to create an aggregate “band enrollment” for that teacher’s band program. This prevented us from needing to identify course levels (e.g., whether the “Concert Band” is the “beginner band”) and allowed the unit of analysis to be each teacher’s program, with multiple individual enrollment counts calculated for each program across several years.
Beyond annual enrollment counts, we also incorporated other observable school and teacher characteristics associated with each music teacher’s ensemble program for our analyses. These additional data were merged from the MSDE and Maryland Department of Labor data warehoused in MLDS. Drawing from a meta-analysis on teacher turnover and job satisfaction (Nguyen et al., 2019) and a conceptual framework for teacher turnover (Nguyen & Springer, 2023), we included observable characteristics that consistently have been associated with job satisfaction and turnover. Specifically, we incorporated teacher sex, whether a teacher was prepared at an in-state or out-of-state postsecondary institution, the total number of student suspensions in the school, the prevalence of chronic student absenteeism within the school, the size of the school expressed in total student enrollment count, the proportion of students eligible for free- or reduced-price lunch through the National School Lunch Program, and the proportion of students of color enrolled in the school. The final analytic data set was unique at the school music program by year level, which allowed us to identify teacher turnover and student enrollment numbers, matched with other teacher and school characteristics.
A key element of our analysis hinged on the designation of teachers as movers, leavers, or stayers in line with the classification used in prior literature. The longitudinal nature of the MLDS data allowed us to observe music teachers across multiple academic years, meaning we could easily discern any particular teacher’s employment status within Maryland from the start of an academic year to the start of a succeeding academic year. That is, teachers’ employment outcomes in analytical year t were based on their observed status at the start of year t + 1. We combined all separate analytical years to maximize the number of observations used in our estimates of the effect of student enrollment on teacher employment status.
Although the reality of teacher turnover disposition is more complex than it appears at first glance, the mover designation is the most straightforward. If a music teacher left the position they held at the start of a particular school year or by the end of that school year, we searched the school workforce data to see if they were observed teaching music in a different Maryland public school the following year. If they were, we designated them as a mover. For example, if Mx. Jones was teaching choir at Old Bay High School at the start of academic year 2015–2016 (year t) and was teaching choir and general music at Old Line State High School at the start of academic year 2016–2017 (year t + 1), we would categorize their outcome for the 2015–2016 analytic year (year t) as ‘‘mover’’.
The leaver situation is somewhat more complicated; by necessity, if we could not locate a teacher departing their position teaching music within a Maryland public school the following year, we designated them as a ‘‘leaver’’. Teachers who remained teaching but left the subject of music in a subsequent year were also designated as leavers. One caveat for teachers designated as leavers is the possibility that a teacher left teaching in a Maryland public school to take a music teaching position in another U.S. state, given that we could not observe out-of-state teacher employment. However, prior research shows that only a small fraction of teachers who move schools cross state boundaries (Goldhaber et al., 2015), suggesting that this possibility likely had a negligible effect on our results. We also designated as leavers those music teachers who left public school teaching for private school employment because we could not observe the specific teaching assignments of private school teachers. We are unconcerned about the impact of this specific situation on our estimates because the number of leaver public school music teachers who were subsequently employed as private school teachers in Maryland was below the 10-observation minimum reporting threshold imposed by MLDS for data privacy.
Finally, although the existing teacher turnover literature does not consider retirement or moving to a different subdiscipline as salient outcomes, these classifications were relevant to our study and needed to be addressed. We, therefore, expanded the outcome classifications to include two additional categories. The first new category identified teachers who remained at the same school but moved to a different music subdiscipline teaching assignment within that school—for example, a teacher may have taught band and choir in 2017–2018 but taught only band in 2018–2019. We designated those teachers remaining at a school with a new music program teaching assignment as “changers.” Our second new category identified those teachers who left their positions due to retirement. To capture this situation more accurately, we designated teachers who left their positions after attaining 25 or more years of teaching experience as “retirees” because 25 years of experience places a teacher at Maryland’s threshold for full benefits from the public teacher pension. Separating out changer and retiree variation removed potential bias in our results that would have resulted by retaining the extant literature’s tripartite classification.
Empirical Approach
To address Research Question 1, we first fit three logistic regression models to examine the relationship between program enrollment trends, teacher, and school characteristics and teacher turnover. In Model 1, we included enrollment trends for the past 4 years, scaled in a decline of 10 students enrolled in the high school music program. In Model 2, we added teacher and school characteristics. In Model 3, we used a two-way fixed effects (TWFE) estimator to account for all the time-invariant characteristics at the music program/school level (e.g., school size, urbanicity) and the school-invariant characteristics at the year level (e.g., annual fluctuation in job market stability). The advantage of a TWFE estimator is that resulting coefficients for enrollment declines are unbiased from all program and time characteristics (e.g., program teacher, the school in which the program is contained, a particularly volatile year of teacher mobility, etc.). Furthermore, we calculated all standard errors using the robust Huber-White sandwich estimator (Angrist & Pischke, 2009). The equation for Model 3, our preferred model of the three, is shown in Equation 1.
In Equation 1, the left-hand side represents the log odds of the probability that a music teacher would attrite, α represents the baseline odds when all predictors are zero in the reference program in the reference (i.e., the intercept), Enrollmentij represents the difference in student enrollment the ith program from the jth school year to the previous school year, and consequently, Enrollmentij-1 represents the enrollment difference from the year prior to the jth year and 2 years prior to the jth year. The remaining lagged Enrollment variables function similarly. MUSICPROGRAMi and YEARi represent the program and year fixed effects. As a generalized linear model, there is no error term in the equation. The jth year is set to the year in which we are observing whether a teacher attrited; because our bandwidth represents rolling time windows across 8 academic years, any music program may be represented in up to 4 jth years with attendant two lagged observations to establish the enrollment trend in the program.
Next, we used multinomial logistic regression, which is particularly well suited for labor market outcomes (Pforr, 2014), to simultaneously estimate the same models for predicting several types of music teacher turnover: staying in the same school but changing teaching responsibilities (changer), staying in music teaching but moving schools (mover), leaving public school music teaching in Maryland (leaver), or retiring out of the teacher workforce (retiree). We used the same set of three models previously described, but this approach provided a more nuanced examination to compare the strength of the relation between enrollment trends and each turnover outcome compared to staying (Long & Freese, 2014). To prevent listwise deletion and improve statistical power in the more complex multinomial models, we limited lagged enrollment changes to 2 previous years rather than 3.
To investigate Research Question 2, we exploited variation in teacher turnover timing and location to conduct a comparative interrupted time series (CITS). A CITS is a strong quasi-experimental design that can yield causal interpretations (Shadish et al., 2002), previously used in the music education literature to examine outcomes such as the effect of No Child Left Behind’s enactment on nationwide music enrollment (Elpus, 2014). Identification of causal effects in CITS relies on several key assumptions (Hallberg et al., 2018). Perhaps most importantly, treatment groups must have comparable control groups that were similar throughout time. In the present analysis, the treatment was teacher turnover. A high school music program, then, was considered for the treatment group if it had an observable window of teacher turnover and for the control group if there was an observable period with no teacher turnover. Additionally, inclusion criteria for treatment programs required a high school music program to have had at least 3 years of observable enrollment trend data prior to a teacher change and 3 years of observable enrollment trend data with the new teacher. Inclusion criteria for control programs required that programs had at least 6 years of observable enrollment trend data without experiencing a teacher change. The next assumption is that there was no instrumentation threat, in other words, the outcome must be measured equivalently at all points in time. We used total music program student enrollment as our outcome and measured enrollment consistently across all points in time, eliminating the instrumentation threat. The final key assumption is no selection threat to validity. That is, the composition of the treatment groups must be consistent in pre- and posttreatment periods. Because our treatment groups (i.e., music programs experiencing teacher turnover) were based within a singular school and the observation period was relatively short, there was likely no selection threat to validity because school composition tends to be stable over short periods of time. We deliberately ended our window of employment observation at the start of the 2019–2020 school year to avoid the history threat to validity that would be introduced by COVID-19 response-related motivations for teachers to leave their job or for students to opt out of music ensemble participation. The empirical model underpinning the CITS analysis is shown in Equation 2.
As an exploratory analysis, in Research Question 3, we used descriptive statistics to learn more about teachers who were observable in the Department of Labor data who left the profession. For this portion of the study, the teacher served as the unit of analysis. In the labor data, we were able to observe both the industries in which former teachers worked and the wages they earned in their new careers. Subsequently, we compared earnings from their final year of teaching and their first year of new employment. Importantly, only individuals employed by organizations that are subject to Maryland unemployment insurance were identifiable in the wage data. This means that a sizable portion of individuals, such as those who were self-employed, those working in the federal government or as federal government contractors (a common employment situation in the state of Maryland), and gig workers such as Uber drivers and Instacart shoppers, among others, are unobservable in the data set.
Results
We organize our results by research question. To comply with data privacy restrictions, all results reported here were reviewed prior to dissemination by MLDS Center staff to limit deductive disclosure risk. Data privacy restrictions for MLDS require that all reported sample sizes be rounded to the nearest 10, all percentages be rounded to the nearest whole number, and any cell with fewer than 10 individual observations be suppressed or combined with other categories.
Descriptive Statistics of Analytic Sample
Our final analytic sample included 750 unique teachers from 580 unique high school ensemble music programs representing 240 unique high schools between academic years 2013 and 2020. In total, there were 4,020 unique (at the music program by year) analytic observations. Supplemental Table S1 included with the online version of this article displays descriptive statistics for the analytic sample. Across years, the average experience of high school music teachers ranged from 12.8 years to 13.6 years. Female music teachers constituted between 35% and 40% of high school music program teachers. People of color constituted 18% of high school music teachers in 2013; this proportion grew steadily through 2020, when 27% of all high school music ensemble programs were taught by a person of color. The rate at which teachers left teaching or moved schools varied considerably from year to year. The year with the lowest rate of attrition was 2018, when only 12% of teachers left their position; the year with the highest rate of attrition was 2013, when 24% of teachers left their position.
Enrollment Declines and Teacher Attrition
Full logistic regression results for the relationship between enrollment declines and teacher attrition are shown in Table 1. In the base model, an enrollment decline of 10 students compared to the previous year was significantly associated with increased attrition risk (OR = 1.354, p < .001). In Model 2, which added several observable teacher and school characteristics, the relation between enrollment decline and attrition weakened somewhat but remained statistically significant (OR = 1.145, p < .05). In Model 3, we used the TWFE estimator to generate estimates of the relation between enrollment decline and teacher attrition that were unbiased from any program-level characteristics or yearly variation. Although TWFE estimators are generally used to generate model-based causal estimates with panel data, recent statistical advances demonstrated that resulting estimates may be biased if model assumptions do not hold (Imai & Kim, 2021); we remain unconvinced that all potential confounders are controlled in our TWFE model due to the complex and multifaceted nature of teacher turnover. In addition to a significant relationship between an enrollment decline from the immediate past year and attrition (OR = 1.395, p < .001), previous years’ enrollment trends were also predictive of attrition. For example, if a teacher left at the end of the 2018 school year, student enrollment declines lagged by 2 and 3 years (i.e., declines in enrollment experienced between 2015–2016 and 2014–2015) were also significantly predictive of that attrition (OR = 1.094, p < .05; OR = 1.08, p < .05, respectively). That is, enrollment declines did not need to occur immediately prior to a teacher leaving their job to be predictive; predictive student enrollment declines could have occurred as many as 3 years prior to the year of teacher attrition. Interpretability of odds ratios can be improved by examining marginal effects, which we show in Table 2. In our preferred model (Model 3), a decline of 10 students in the music program student enrollment was significantly related to a 7.9 percentage-point increase in attrition risk (p < .001). Lagged enrollment declines from 2 and 3 years prior were significantly related to a 2.2 and 1.8 percentage-point increase in attrition risk, respectively (p < .05).
Logistic Regression Models for Teacher Attrition.
Note. Enrollment deltas are scaled in declines of 10s of students. Coefficients are reported as odds ratios. Robust standard errors are in parentheses.
p < .05. ***p < .001.
Marginal Effects of Program Student Enrollment Declines on Attrition Risk.
Note. Enrollment deltas are scaled in declines of 10s of students. ME = marginal effect.
p < .05. **p < .01. ***p < .001.
Multinomial logistic regression provided a more nuanced examination of the relationship between teacher attrition and music program student enrollment declines by differentiating between changers (those who remained in the same school but changed responsibilities), movers (those who moved schools), leavers (those who left public school music teaching in Maryland), and retirees (leavers with more than 25 years of experience who were presumed to have retired). Multinomial logistic regression results are included in online Supplemental Table S2, and marginal effects of Model 1 are included in Table 3. Although our preferred model was Model 3, our use of TWFE multinomial logistic regression precluded marginal effect estimations (Pforr, 2014) because in a fixed-effect multinomial logistic regression model, marginal effects “cannot be estimated because the unobserved heterogeneity vector is not estimated” (Pforr, 2014, p. 851). Fortunately, the relative risk ratios estimated in Model 1 are comparable with the estimated odds ratios in Model 3, so we present marginal effects from Model 1 to improve the interpretability of odds ratios. For example, the relative risk ratio (RRR) for an enrollment decline of 10 students in our preferred model for the outcome mover (RRR = 1.39, p < .001) at a surface appears comparable to the outcome leaver (RRR = 1.42, p < .001). However, the multiplicative nature of odds ratios can obscure their true effect, particularly when the baseline probability is low. The marginal effects demonstrate that an enrollment decline of 10 students was significantly associated with a 0.3 percentage-point increase in risk for moving but a much higher 5.2 percentage-point increase in risk for leaving. Conversely, a decline of 10 students was significantly associated with a lower propensity for staying (marginal effect = .055, p < .001). Multinomial logistic regression results affirmed and clarified results from the logistic regression model that when a high school ensemble music program experiences a decline in student enrollment, the teacher is more likely to leave the profession, not just their present position.
Marginal Effects of Program Student Enrollment Declines on Labor Market Decisions.
Note. ME = marginal effect.
p < .05. ***p < .001.
The Impact of Teacher Turnover on Program Enrollment
The CITS analysis revealed no significant effect of a new teacher on future music program student enrollment (β = −1.27, p > .05), that is, when a new teacher began in a music ensemble program, there tended to be no significant change in enrollment trend. Any preexisting secular trend (i.e., the trend absent any intervention), whether increasing or decreasing, tended to remain the same after a personnel change of teacher. We also conducted subpopulation analyses based on the years of experience of the leaving teacher: one analysis limited to teachers with 10 or fewer years of experience (β = −7.87, p > .05) and one analysis on teachers with more than 10 years of experience (β = 1.78, p > .05). When a newer teacher left their position, enrollment in the subsequent years declined; when a more veteran teacher left their position, enrollment in subsequent years grew a modest amount. Full results are included in online Supplemental Table S3. One important limitation to note is that our strict inclusion criteria reduced the number of programs included in all CITS analyses; thus, these analyses had a relatively low statistical power to detect small effects. Although none of these analyses yielded coefficient estimates that were statistically significant, the magnitude of difference and change in sign between the subpopulation analyses warrant additional investigation into this topic.
Post-teaching Careers of Former High School Music Teachers
Of the 720 unique teachers observed leaving the public music teacher workforce, 290 were observable in the labor force data. The demographic composition of those observed in the labor force data was approximately the same as the composition of all leavers, suggesting our subset of observable former music teachers is likely representative of all former music teachers. The 430 leavers who were not observable in the labor force data may have been unobservable due to unemployment, self-employment, employment in another state (either as a music teacher or in another profession), or employment in Maryland in any organization that is not subject to Maryland Unemployment Insurance, such as the federal government or as one of its contractors. Of leavers who were observable, the overwhelming majority (98%) continued to work in the education industry in some capacity. Specific places of employment and job titles are not observable in MLDS data, but the “education industry” code spans a wide range of nonteaching careers, such as school administrator, support staff, counselor, district-wide resource teacher, or music supervisor. The remaining 2% of leavers were observed working in various other industries, including waste management and remediation services, finance and insurance, and wholesale trade.
When comparing leaver wages from teaching and their next nonteaching career, 67% of observed leavers had higher earnings in their new careers. It is also plausible that leavers with employment unobservable in MLDS also had higher earnings; for example, in the time period we examined, a federal contractor would tend to make more than a teacher. In their teaching careers, leavers had an average salary of $68,364.74 and a median salary of $64,781.50. In comparison to new career earnings, the average difference was $4,260.05 higher than teacher earnings, and the median difference was $8,219.00. Finally, 25% of observable leavers—both those who remained in the education industry and those who did not—experienced a wage increase of $21,865.00 or more. Although it is impossible to say with certainty, it is a reasonable conjecture that the opportunity for higher wages and earnings contributed to music teachers’ decision to leave the profession.
Discussion
This article is the first in the field to empirically investigate the possibility that student enrollment declines may lead a music teacher to leave their job. In line with the conventional wisdom of the profession, declining student enrollments in a secondary music ensemble program was a salient predictor of music teacher attrition. This basic result held across all of our logistic and multinomial logistic regression models, suggesting that the result is robust to various possible confounds. In our comparative time-series analysis, we were unable to detect any significant effects of music teacher turnover on future student enrollment trends. Given the pattern of our results, we first discuss the results of Research Question 1 and then consider the study results taken together as we discuss implications for school administrators, music educators, music teacher education, and future directions for music education research.
Within the broader education literature (e.g., Boyd et al., 2010), teacher effectiveness has consistently been demonstrated as a strong predictor of teacher attrition. However, music teacher effectiveness is neither concretely defined nor easy to measure. Returning to Fitzpatrick’s (2011) concept of thriving and struggling music programs, music programs with declining enrollments would be characterized as struggling. The conventional wisdom we sought to empirically test would suggest that on average, less effective music teachers would be more likely to teach in struggling music programs with declining enrollments. Thus, declining enrollments could serve as a proxy indicator for teacher effectiveness and consequently, would be predictive of attrition. The results of our analyses demonstrated that declining enrollment does indeed predict teacher turnover, supporting our theory of action and lending some empirical evidence to the notion that teacher success, if not effectiveness, may be linked to increased student enrollment.
A limitation of extrapolating from our analysis is that our theory of action hinges on a strong assumption that less effective teachers are less able to recruit and retain students compared to more effective teachers. However, student enrollment may decline for reasons other than teacher effectiveness, including factors completely outside of a music teacher’s locus of control, such as lack of administrative support, scheduling challenges, or changes in the success of the “feeder” program at the middle school. To address this limitation arising from the logic underpinning this study, we provide three alternative theories of action that warrant consideration for research and the broader music education profession.
One alternative explanation for declining enrollment unrelated to teacher effectiveness could be a lack of administrative and financial structures to support recruitment processes. For example, a teacher may not have a planning period that permits travel to a middle school. Because the transition between middle school and high school is a large point of music student attrition (Elpus, 2022b; Evans et al., 2013; Tucker & Winsler, 2022), high school music teachers may increase their enrollment through middle school visits that build positive relationships with potential students prior to course registration. Perhaps a high school teacher wanted to provide buses to bring middle school students to the high school campus, but their administration did not approve the necessary funding. In situations such as these, music student enrollment could then decline not because of low teacher effectiveness but due to a lack of administrative support. Because administrative support is strongly predictive of music teacher workplace satisfaction and retention (Baker, 2007; Gardner, 2010; Hancock, 2008; Killian & Baker, 2006), this alternative theory of action could also be a viable explanation for the pattern of results we find: that declining enrollment predicts teacher attrition.
A second alternative explanation is that outside factors could cause a drop in enrollment, such as declining middle school enrollment or, absent a trend, year-to-year variation in middle school music enrollment. The reason for lower high school enrollment may not matter. The real impact felt by the high school music teacher is the same: Declining enrollments may lead to higher levels of stress and lower workplace satisfaction. These stressors could include lack of ensemble balance, the inability to perform certain compositions due to the absence of students in certain instruments or voice parts, or the internalized perception that colleagues see declining enrollments as an indicator of struggle. From an employment perspective, enrollment declines may also lead to reduced hours or full-time equivalency because many music teachers’ teaching assignments are directly related to the number of students enrolled in the program. Higher stress and part-time status may then lead to lower workplace satisfaction and burnout (McLain, 2005) and consequently, attrition.
Finally, we offer an alternative explanation that flips the temporal ordering. Although attrition is ultimately observed at the end of the school year, teachers may contemplate these decisions well in advance due to any number of personal or professional reasons. Under this theory, it is conceivable that a teacher may decide to resign from their position 1 or 2 years prior to actually leaving. During this period after the decision to leave has been made but before it is enacted, the teacher may not invest as much time or effort into recruiting students for the music program, resulting in declining program enrollment in the years preceding attrition. This plausible theory explaining our pattern of results supports the conventional wisdom that higher enrollment is linked with greater teacher effectiveness in a nuanced way. This theory links student enrollment not with the capacity to be a highly effective music teacher but with the enactment of being a highly effective music teacher.
In considering our exploratory analysis, we found a stark difference between teacher salaries and earnings from employment after leaving music teaching. Earnings constitute one aspect of workplace satisfaction, and we found that many music teachers who left indeed received higher wages in their next career. Comparatively low salaries could have contributed to lower workplace satisfaction and the decision to leave teaching. Although there is no singular cause for attrition among teachers, researchers have consistently found that workplace satisfaction strongly predicts attrition or attrition intention among all teachers broadly (Nguyen, 2021; Nguyen et al., 2019; Nguyen & Springer, 2023) and music teachers specifically (Baker, 2007; Gardner, 2010; Hancock, 2008; Killian & Baker, 2006).
Implications for Music Educators and School Administrators
No matter which theory of action accurately explains our pattern of results, declining student enrollment was a prescient indicator of attrition among high school ensemble teachers and may be a useful metric for music administrators to monitor. Because each local context is unique, any one of these theories of action could ostensibly be at play. School administrators and local school systems should then consider each program uniquely to best understand the barriers to student enrollment within a high school ensemble music program. Because teacher turnover is costly (National Commission on Teaching and America’s Future, 2017) and contributes to a negative school climate (Carver-Thomas & Darling-Hammond, 2019), administrators have a vested interest in retaining teachers.
Declining student enrollment may serve as an easily surveilled early warning indicator that a music teacher needs some form of support or intervention. Importantly, the declining enrollment may be an observable symptom of a teacher’s challenge or struggle with an element of music teaching that can be remedied before leading to teacher attrition. For example, music teachers in need of intervention may benefit from mentorship to improve teacher effectiveness and student retention. Mentorship opportunities could be facilitated by release time to attend professional development conferences. Support could also be provided directly by administrators asking the teachers how to best serve their needs. Music teachers are likely to perceive administrators as more supportive if administrators take proactive measures to address their teachers’ concerns directly, particularly regarding program enrollment and teaching effectiveness. Given the mythos surrounding program size and many music teachers’ efforts to recruit and expand enrollment, a shared goal between teacher and administrator could help foster positive relationships, increase perceived support, recruit more music students, and reduce attrition risk.
Music educators themselves may also benefit from tracking their own program enrollment trends when implementing new teaching strategies and other program changes over time. Although the direct success of any such change may not be readily observable and measurable, greater student enrollments may serve as a useful proxy indicator for successful implementation and higher teacher effectiveness. The success of varying teaching strategies, such as placing increased emphasis on student creative musicianship or peer assessment and feedback, then have the potential to be evaluated via enrollment trends over time. If program enrollment has grown after several years of implementing these teaching strategies, our theory of action would suggest the development and implementation of these teaching strategies was effective.
Music educators should also be aware of the direct pipeline for recruiting music students into their program. The plethora of practitioner literature regarding recruitment and retention of music students (e.g., Brinson & Demorest, 2012; Colwell et al., 2017; Hash, 2022; Sandene, 1994) provides pertinent strategies for recruiting students. Structural systems, such as course scheduling and elective options at the secondary level, must also be considered. Investing time in recruitment and working with administrators to bolster recruitment efforts may improve job satisfaction and lower attrition risk. Additionally, high school music educators, particularly those with program enrollment near or below the threshold for their local school system to reduce their full-time equivalence, could consider offering a beginning-level class. By creating an entry point to these ensemble music courses at the high school level, teachers could accomplish two things. First, teachers would greatly increase access to these music education courses for all students who did not enroll at the point when music first became elective. Prior research (e.g., Miller, 2023b) has shown that middle school music participation is one of the most salient predictors of high school participation, potentially due to the lack of beginning-level classes. Second, the increased enrollment from beginning students could mitigate the risk of reduction in full-time equivalence and subsequently reduce the likelihood of attrition. Beginning-level classes for high school students would augment current recruitment pathways for ensemble music programs, providing ensemble music opportunities to a broader school population and safeguarding against a reduction in full-time equivalency.
One common experience shared among young music teachers taking positions formerly held by long-tenured veteran music teachers is the stress they feel in maintaining the historical reputation of the program. Declining enrollment trends as the new teacher remains in the position might compound these stressors because the new teacher may feel as though they are not “living up” to the expectations of their predecessor or the school community. Our results suggest, however, that any troubling enrollment trends immediately following a music teacher personnel change are not likely to persist.
Implications for Music Teacher Education
Maryland and many other states have recently experienced or are currently experiencing a shortage of music teachers (Hash, 2021; Miller, 2023a). Because music teacher educators are tasked with the recruitment and preparation of music teachers for elementary and secondary schools, helping preservice music teachers cultivate a skill set for career longevity is vital to mitigate shortages and halt the revolving door of teacher turnover. Although the exact mechanism through which student enrollment predicts teacher attrition has yet to be fully explicated, the results of the present study clearly delineate the importance of successful student recruitment and high school ensemble music teacher longevity. Music education curricula should emphasize recruitment strategies, and preservice teachers completing internship experiences should work with their cooperating teachers for recruitment activities when possible. Strong recruitment skills will not cease all attrition, but ensuring preservice teachers develop this skill set and understand the importance of recruiting for high school music programs will reduce attrition risk substantively.
Music teacher educators can consider program growth when identifying possible teachers with whom to partner for preservice internships. Fitzpatrick (2011) characterized one facet of thriving programs as size; our results build on this conceptualization as we posit a link between teacher effectiveness and program growth rather than solely program size. Music program enrollment is likely related to total school enrollment (Thomas et al., 2013); larger school enrollments tend to yield greater absolute numbers of music students, although this relationship is likely strongest in affluent, suburban high schools (Elpus, 2022a). Music teacher educators, then, might consider a music teacher whose program has grown, independently of the absolute size of the program, as they explore possible student-teacher mentors. This consideration might help identify possible sites for internship experiences in multiple diverse contexts rather than relying on a relatively homogenous set of large ensemble programs as internship sites. Furthermore, because we found that growing student enrollments predicted a greater likelihood of a teacher staying in their position, fostering partnerships with teachers experiencing enrollment growth may lead to longer lasting partnerships.
Implications for Future Research
Our results give rise to many possible explanations. Choosing the best theory elucidating the empirical reality from among the candidate explanations requires further research with some refinements in method. We lacked direct measures of teacher effectiveness and job satisfaction in our data set, so additional research into these constructs and their relationship with music teacher turnover would be able to affirm, extend, or disconfirm these various theories. Given the complex nature of these constructs, qualitative inquiry that extends the literature on what makes a music program “thriving” or “struggling,” centers the music teacher, and explores issues such as workplace satisfaction, recruitment strategies, program growth over time, and administrative support would provide the necessary context for our findings.
In the present study, we were the first to directly link declining student enrollment in high school music ensemble programs over time to music teachers’ exit from the profession. Because our data source was constrained to the state of Maryland, replication studies and additional research in other contexts and other locales are necessary to determine the extent to which these results scale. Furthermore, we ended data analyses prior to the onset of the COVID-19 pandemic, which may impact future findings. A natural direction for future research would be to replicate our study using the SLDSs of other states. Although the data availability and access requirements vary by state, the present study demonstrates that SLDSs are rich data sources rife with opportunities to answer research questions that otherwise escape music education researchers. Similar investigations or investigations on other topics using other SLDS data would be a welcome addition to music education scholarship.
Supplemental Material
sj-pdf-1-jrm-10.1177_00224294231206098 – Supplemental material for Do Declining Enrollments Predict Teacher Turnover in Music?
Supplemental material, sj-pdf-1-jrm-10.1177_00224294231206098 for Do Declining Enrollments Predict Teacher Turnover in Music? by Kenneth Elpus and David S. Miller in Journal of Research in Music Education
Footnotes
Acknowledgments and Data Availability Statement
Data for this study were drawn from the Maryland Longitudinal Data System (MLDS) Center 
. We appreciate the feedback received from the Maryland Longitudinal Data System Center and its stakeholder partners. All opinions are the authors’ and do not represent the opinion of the MLDS Center or its partner agencies.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study is a project of the Music & Arts Education Data Lab (), a National Endowment for the Arts Research Lab. The Music & Arts Education Data Lab is supported by Grant 1891756-38-22 from the National Endowment for the Arts. The opinions expressed in this publication are those of the authors and do not represent the views of the National Endowment for the Arts or the NEA’s Office of Research and Analysis. The Arts Endowment does not guarantee the accuracy or completeness of the information included in these materials and is not responsible for any consequences of its use.
Author Biographies
). He is also a researcher with the Research Branch of the Maryland Longitudinal Data System Center. His research interests include music in education policy, issues of demography and representation among music students and music teachers, and the social and academic consequents of music and arts education for K–12 students.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
