Retail employee turnover and turnover destinations – the role of human capital

ABSTRACT

This paper assesses the role of human capital in labor turnover and turnover destinations of full-time retail employees. We use register data that encompasses the full population of Swedes above the age of 16 and follow the career paths of those that have, at one point in time between 2002 and 2018, worked full-time in retail. We use logit- and multinomial logit-estimations to assess the role of firm-specific (proxied by establishment tenure) and worker-specific human capital (proxied by industry tenure, formal education, retail education, and occupational complexity) in the propensity to quit a retail establishment and the retail sector. Results indicate that establishment tenure, industry tenure, retail education, and occupational complexity decrease the probability of quitting, while formal education has the opposite effect. Moreover, we find that industry tenure, retail education, and occupational complexity increase the probability of staying in the retail sector.

KEYWORDS:

• Retail
• career paths
• Sweden
• panel data
• human capital

Introduction

The retail sector is a vital part of many mature economies, and Sweden is no exception, where it employs 5% of the labour force.¹ The sector’s employment capacity is particularly important in promoting the integration of young (unemployed) individuals’ entry into the Swedish labour market (e.g. Daunfeldt and Fergin-Wennberg 2016). According to the Swedish Retail and Wholesale Council,² many of those who are currently employed started their careers in either the retail or the wholesale sector, which further signals the importance of the sector as a whole.

One of the major challenges faced by the sector is high labour turnover (Han, Håkansson, and Lundmark 2019; Knee 2002) and in Sweden, approximately one-third of all employees exit their establishment on a yearly basis.³ This entails costs for the firms in the form of productivity losses, short staff-spells, as well as a loss of team morale (Kuhn and Yu 2021). The rapid changes taking place in the retail sector, such as automation and competition from e-commerce channels – the latter which has intensified during the COVID-19 pandemic – are extensions of the structural changes in the economy, which increase the importance of competent personnel in several aspects. As argued by Mende and Noble (2019), retail staff at the frontline are becoming more important from both the supply and demand side. For instance, with increased access to product information, consumers’ knowledge of a product prior to retail contact creates additional demand on the competence of the front-line personnel (Hochstein et al. 2019, 2021). Furthermore, the COVID-19 pandemic requires a reconfiguration of traditional offline consumer interactions to safer digital interactions (Roggeveen and Sethuraman 2020), which increases the pressure on the competencies of retail employees. The ability of an establishment and the industry to retain competent employees is thereby important to remain in an increasingly competitive market (Brush and Chaganti 1999).

This study aims to assess the role of human capital in an individual’s propensity to quit a retail establishment and the retail industry. We focus on two types of human capital: firm-specific⁴ and worker-specific. The former is captured as firm tenure, and the latter as industry tenure, formal education, specific retail education, and occupational complexity. Thereby, we can inform establishments and firms about the characteristics of employees that are at risk of quitting, and we can inform retail business associations of the kind of human capital development that should be promoted. Our analysis makes several contributions. Factors explaining labor turnover in the retail sector have been well studied; however, of the conducted studies, many are case studies (e.g. Dube, Giuliano, and Leonard 2019; Hendrie 2004; Salleh, Nair, and Harun 2012), where the results might not be generalizable. In this paper, we use the whole population of retail employees and can thus draw inferences from a larger group. Furthermore, most studies that focus on human capital in retail or service sectors are cross-sectional (e.g. Salleh, Nair, and Harun 2012; Stamolampros et al. 2019), and they are thus unable to capture job transitions within and across sectors. The few studies that do consider repeated observations (e.g. Dube, Giuliano, and Leonard 2019) are based on older datasets from before the 2000s. With the rise of e-commerce, the retail sector has seen a drastic structural change since that time and consequently, there is a need to analyse the supply of skills in the retail sector using more current data. The decision to leave an establishment further implies a turnover destination, for instance, in the form of employment in a new sector. There is currently an overall lack of studies analysing turnover destinations in general (Zimmerman, Swider, and Arthur 2020). We contribute to this line of research by focusing on the role of human capital for turnover destinations within or outside retail. Moreover, there is a lack of studies that assess the relevance of factors that could be influenced by policy measures, such as the level and type of education and industry experience, as well as the characteristics of the work tasks, which is the focus of this paper.

We use detailed register data to follow all individuals above the age of 16 in Sweden over the period 2002 to 2018. We focus on employees that were working full-time in the retail sector in time t and assess how human capital may induce a potential change in t + 1 by employing a set of logit estimations. Our binary dependent variable indicates if the employees in year t + 1 (i) stay in their job or quit; (ii) whether they quit their job but take on a new job in the retail sector; or (iii) whether they quit their job and take on a new job in a different sector. We use the establishment as our unit of analysis.

In this paper, we study what we argue to be mainly voluntary labor turnover. The fact that we study full-time employees – which implies that a job separation involves costs to the firm – in combination with the exceptionally strong position of workers’ rights on the Swedish labor market – makes it very difficult to fire employees in general, and full-time employees in particular – supports this argumentation.

Our results indicate that firm-specific human capital in the form of establishment tenure decreases the probability of quitting the establishment, while worker-specific human capital in the form of formal education increases the probability of quitting the establishment. Worker-specific human capital in the form of industry experience, retail education, and occupational complexity,⁵ however, decreases the probability of quitting the establishment. Of those who do quit an establishment, we find that establishment tenure decreases the probability of staying in the retail sector, while industry experience, retail education, and occupational complexity increase the probability of staying in the retail sector.

The next section presents an overview of previous studies addressing labor turnover. Section 3 describes the data, and section 4 describes the empirical design. Section 5 presents and discusses the empirical findings and section 6 concludes.

Human capital and labor mobility

According to Porter and Steers (1973), the underlying factors of labor turnover can be broadly nested under four categories: individual, working environment, firm, and type of tasks performed by the employee in a firm. In this paper, we focus on individual factors in the form of human capital. Human capital – both firm-specific and worker-specific – is related to the probability that an employee decides to leave a firm (Becker 1964). Firm-specific human capital is valuable for the establishment specifically, in the form of on-the-job training and specialized learning-by-doing and cannot be transferred across firms. Worker-specific human capital is gained through formal education and industry experience and can be applied across firms.

Following studies on quit behaviour (e.g. Lévy-Garboua, Montmarquette, and Simonnet 2007), we use a wealth maximization approach to model the propensity of an individual to leave employment in a firm and/or a sector as a function of expected benefits and costs. The decision to stay or leave depends on the expected value of future benefits of staying, $E_{\mathit{it}}{V}_{it}^{\hspace{0.17em}}$, relative to the expected value of pursuing other alternatives (OA), $E_{\mathit{it}}{V}_{it}^{OA}$, minus the cost of switching jobs, $C_{\mathit{it}}$. The individual will leave an employment/sector if:

(1) $E_{\mathit{it}}{V}_{\mathit{it}}-\left(E_{\mathit{it}}{V}_{it}^{OA}-C_{\mathit{it}}\right)<0$(1)

Which rewritten formulates,

(2) $C_{\mathit{it}}>\left(E_{\mathit{it}}{V}_{it}^{OA}-E_{\mathit{it}}{V}_{\mathit{it}}\right)$(2)

Thus,

(3) $\mathit{Prob}\left(\mathit{Quit}=1\right)=\mathit{Prob}\left(E_{\mathit{it}}{V}_{\mathit{it}}\le E_{\mathit{it}}{V}_{it}^{OA}-C_{\mathit{it}}\right)$(3)

Several factors have a positive effect on the expected value of remaining in the current workplace ($E_{\mathit{it}}{V}_{\mathit{it}}$). As firm-specific human capital is valued higher by an individual’s current employer than by other employers, the individual has a higher incentive to remain in the workplace than to leave. Another argument, by Lévy-Garboua et al. (2007) is that as firm tenure increases (i.e. firm-specific human capital), a worker adapts and becomes used to certain negative traits of the job, which indicates that tenure may decrease the propensity to quit. Firm tenure is found in previous studies to have a negative relationship with the propensity to quit in general (Bal, De Cooman, and Mol 2013; Olsen, Sverdrup, and Kalleberg 2019), and in service industries in particular (for instance, Arndt, Arnold, and Landry 2006; Hendrie 2004; Stamolampros et al. 2019).

Factors that may affect the expected value of seeking alternative opportunities, $E_{\mathit{it}}{V}_{it}^{OA}$, are worker-specific human capital in the form of formal education. Individuals with more formal education have more labor market opportunities and are therefore more likely to quit a job. Another form of worker-specific human capital that has been found to increase the opportunities for alternative employment opportunities is industry tenure (e.g. Miller et al., 2021; Van der Heijden et al. 2018). Becker (1964) argues that industry-specific human capital, in a manner analogous to firm-specific human capital, may increase the value of remaining in the industry. Thus, industry-specific human capital may increase the value of staying in the establishment as well as the value of finding alternative employment in a different establishment in the same sector. Moreover, as argued by Kiazad et al. (2015) this human capital can be seen as a specific investment made by the worker which he or she would be unwilling to give up. Thus, the expected value of pursuing an alternative career in a different sector may decrease with this type of capital. Gable, Hollon and Dangello (1984) analyse managers in US retail establishments and find that previous experience in the same industry increases the probability of staying on at the establishment. Min (2007) finds similar results in a study on the wholesale sector.

Lastly, we have factors that affect the cost of quitting, $C_{\mathit{it}}.$These factors include marital status and family situation, which may affect the value that an individual ascribes to stability. However, studies of the retail sector, e.g. Broadbridge (2007), report that work-family conflicts as being one of the perceived barriers to advancement and Good, Page and Young (1996) show that having a spouse or family increases the propensity to quit. This conflict can be further exacerbated by unpredictable scheduling, as found by Choper, Schneider and Harknett (2019). Other costs of quitting may be related to a sense of morality. In the study by Maertz and Campion (2004), individuals describe their motivation to stay on at a job as due to moral behaviour, where staying is associated with loyalty to the firm. However, loyalty has been found to extend beyond the firm. In a study by (Foster, Whysall, and Harris 2008), of retail employees, one of the more pronounced targets of loyalty involved the store with its colleagues and its interior, which was described by some as a second home. Last, quitting a job may also entail a loss of seniority. Using the construct of on-the-job embeddedness, an employee’s intention to stay is a function of the fit to the firm, his/her links to others at the firm, and the sacrifice or the cost of leaving their job such as personal losses, financial cost, and other related benefits, loss of job stability and career advancements, and other accrued benefits (Mitchell et al. 2001; Sender, Rutishauser, and Staffelbach 2018). Consequently, the length of an individual’s previous employment may be correlated with the perceived cost of quitting the current job.

Data and variables

Using limited public access data from Statistics Sweden with detailed information on all individuals over the age of 16 who are residing in Sweden and data on all active establishments – those that pay value-added tax each year – we create a large employee-employer matched dataset that covers the period between 2002 and 2018. Establishments in the retail sector are identified using the standard industry classification codes (SIC).⁶ The unit of observation in our analysis is the establishment, and we link the individuals to their respective establishments. In our data, we have establishments that are part of the same firm, i.e. defined as chains, and establishments that are independent. We include indicator variables to control for the different types in the estimations. The dataset consists of all individuals that in time t are employed full-time in the retail sector, and their employment status in t + 1. To capture individuals who are employed full-time, we analyse only individuals with an annual wage that is above the annual minimum wage in the retail sector,⁷ which in 2006 was calculated to be 170,000 SEK (~16,000 EUR).⁸ In the sample, we have included all individuals that have worked at least one-year full-time in retail between the years 2002 and 2018. The reason to focus on full-time employees is that any job separation among these is likely to voluntary, for two reasons. The first is that these are the employees whose job separation involves, for the firms, the highest costs in terms of sunk recruitment and introduction costs. The second is due to the characteristics of the labor market regulations. The Swedish labor market is characterized by a strong protection of worker’s rights, and it is one of the top 15 countries with the strictest employment protection (OECD.Stat 2019). Furthermore, it is in the top of countries with regard to work–life balance (OECD.Stat 2021). This in combination with strict rules on the work environment (The Swedish government 1977; 1982a 1982b) makes a full-time job separation likely to be a voluntary decision and not one which is driven by necessity. Part-time workers are also covered by these laws; however, as we cannot identify the type of contract that the employee has, we do not know if it is a part-time employment that is permanent or not. However, by limiting the population to full-time workers we can reduce the risk of including non-permanent employees and whose job separations may be driven by necessity (for instance, students working extra during the semesters will eventually finish their education and therefore also their part-time employment) rather than opportunity. However, part-time employees are essential for the retail firms which can better adjust production for seasonal variations. They also make up a considerable part of the employment in the trade sector in Sweden, 31% in 2019, which is higher than the EU 27 average of 21% but lower than that of the UK of 37% and the Netherlands of 55% (Eurostat 2023).⁹

Variables

Dependent variables: To capture the change in labor market status, we construct several binary dependent variables that record changes in status between one year and the next. The variable indicating a job separation (Quit), is equal to 1 if the employee, at time t, works at a retail establishment and in t + 1 quits this job. In the literature on voluntary job separation, intention to quit is sometimes a preferred predictor of turnover over actual turnover behaviour (e.g. Bertelli 2007; Pitts, Marvel, and Fernandez 2011). In our data, we can only observe the actual job turnover behaviour, and while we cannot know whether this was in fact voluntary, turnover of full-time employees – our studied population – is particularly costly to an establishment due to sunk recruitment and introduction costs and foregone future productivity. In combination with the strict regulations protecting job security, we may therefore assume that the majority of job separations are initiated by the employees. We define a job separation as an employee leaving an establishment, but not necessarily the firm. The argument for basing the analysis on the establishment level is that the risk and the cost of losing an employee lie primarily at the establishment level, rather than at the firm level. Hence, the analysis is also applicable to firms with multiple establishments. In the second step, we analyse whether an employee who has quit his/her job: (i) continues as an employee in the retail sector (Stayed ind.); (ii) switches to a different sector (Left ind.); or (iii) becomes an entrepreneur (Entrepreneur). We also follow individuals who quit their jobs and become unemployed or become full-time or part-time students (Unemployed/studying). We exclude job separations that coincided with the closure/merger or acquisition of an establishment or the death or retirement of an individual.

Individual human capital: Due to the construction of the data, we can only capture the establishment tenure and industry tenure as continuous variables up to 17 years measured from 2002–2018 (Est. tenure; Ind. tenure). It is not possible to discern if individuals with 17 years of tenure have 17 years or more, and thus we use a dummy to indicate this (Est. tenure >16 years; Ind. tenure >16 years). We measure the level of formal education categorized into six different levels (Education). We also include a dummy variable that indicates whether an individual has been enrolled in an upper secondary school trade education (Retail ed.). In Sweden, a vocational preparatory education such as that offered at Swedish upper secondary level schools involves a combination of vocational education with practical on-the-job training (as of 2010, 15 weeks of training is a guaranteed minimum (Ministry of Education 2010)). As such one may view it as an occupational investment by the individual. Such investments have been found to increase the opportunity cost of changing careers (e.g. Zimmerman, Swider, and Arthur 2020); therefore, it may decrease the propensity to quit. However, as it also increases mobility within a specific career, it may have a positive correlation with quitting, and therefore the expected effect of this variable is inconclusive. Lastly, we include the level of complexity in the job tasks, to capture yet another worker-specific type of human capital. A high degree of complexity in the job tasks has been found to have a negative correlation with turnover intention in for instance the banking industry (Falahat, Kit, and Min 2019) and related indicators of turnover in the service sector (e.g. Salleh, Nair, and Harun 2012). To capture the complexity of work tasks, we include an ordinal variable to measure occupational skill (Occup. skill). The variable is based on the occupational code that is assigned to each employee and originates from Statistics Sweden. From the occupational code, it is possible to proxy and infer what types of work tasks an employee normally conducts. The variable that captures occupational skills contains four levels, which are classified according to the type of tasks performed given the occupation, and whether the tasks require formal education, experience, or training. The lowest level involves the least complex tasks, such as non-technical tasks such as cleaning, and the highest level involves tasks such as staff responsibility. These categories are based on international standardization by Statistics Sweden (2001, 2012).

Control variables: The ‘inequity theory’ suggests that employees who perceive that they are unfairly compensated for their work relative to their colleagues may be prone to leave the establishment (Adams 1963). Consequently, perceived inequity in wages increases labor turnover (Booth and Hamer 2007; Dube, Giuliano, and Leonard 2019). We, therefore, include relative wage earnings in the establishment (Relative earnings, est.) which is calculated as the log of an individual’s annual wage minus the log of the average annual wage of workers with the same occupational code. As also wage relative to the rest of the region influences the probability of quitting (e.g. Galizzi and Lang 1998), we include a variable that captures the individual’s wage relative to the labor market region’s average wage for the same occupational code (Relative wage, LA-region).

Furthermore, we control for individuals’ family/relationship status by including variables that indicate: if he/she is married/in a partnership or cohabiting (Married); has children living at home (Children), his/her age (Age cat.) (as a categorical variable, since it is highly correlated with the tenure-variables), gender (Female), and country/region of origin (Region of birth).

We also control for establishment characteristics: The size of the establishment (Est. size), which may affect opportunities to advance and, hence, to stay in the establishment – as found in studies on employee turnover in general (e.g. Schmidt et al. 2018), and retail employment in particular (e.g. Min 2007).¹⁰ We also include a variable that indicates whether the establishment is part of a chain (Chain) indicating that the establishment belongs to a firm with multiple establishments. We further include a variable that captures profitability, in the form of a firm’s operating profit divided by net turnover (Marginal return).

Lastly, we also control for regional characteristics: a variable for the size of retail employment (full-time employees) in the labor market region (Retail region), which captures alternative opportunities in the region, and an ordinal categorical variable that indicates the centrality of the municipality (Central), ranging from the least central, category 1 (rural and very remote), to the most central, category 6 (metropolitan), according to the definition by The Swedish Agency for Growth Policy Analysis (2014). Table A1 in the appendix shows the descriptive statistics for our dependent and independent variables.

Empirical design

We model the factors affecting an individual’s decision to quit and, conditional on quitting, the subsequent employment outcomes. However, as we only observe individuals working in the retail sector and individuals who quit their jobs, the outcome variables are truncated, which could warrant the use of a bias correction model. However, in our case, the subsamples – individuals working in the retail sector and individuals who quit their jobs – are the populations on which we wish to generalize, therefore we use traditional dichotomous choice modelling.

In the following text, we describe the different outcomes and how they are empirically estimated.

Quit

As the dependent variable (remain at a job or quit) is binary, we conduct a logit estimation. The logit estimation captures the cumulative probability that an employee quits and compares it to the probability of remaining at a job. The following equation shows the logit model to be estimated:

(4) $\mathit{P}\left(Q_{\mathit{it}+1}=1\right)=\frac{e^{\alpha +{\beta }_1{\mathit{X}}_{\mathit{it}}+{\beta }_2{\mathit{X}}_{\mathit{ft}}+{\beta }_3{\mathit{X}}_{\mathit{rt}}+\mathit{year}}}{1+e^{\mathit{\alpha }+{\mathit{\beta }}_{1\mathit{t}}{X}_{\mathit{it}}+{\beta }_2{X}_{\mathit{ft}}+{\beta }_3{X}_{\mathit{rt}}+\mathit{year}}}$(4)

where $Q_{\mathit{it}+1}$ is the dependent variable that is equal to 1 if individual i in period t + 1 decides to quit his/her job. The explanatory variables are factors that have been established by the literature to affect voluntary labor turnover in the retail sector, as explained in the previous section. ${\mathit{X}}_{\mathit{it}},\mathit{\hspace{0.17em}}{\mathit{X}}_{\mathit{ft}}$, and ${\mathit{X}}_{\mathit{rt}}$ are vectors of our explanatory variables for individual i, establishment f, and region r, at time t.¹¹ Last, we have included a time-dummy, year.

Paths after quitting

For the second step, we use a multinomial logit model:

(5) $\left(Y_{\mathit{it}+1}=\mathit{K}\right)=\frac{{\mathit{e}}^{{\mathit{\beta }}_{1\mathit{K}}{X}_{\mathit{it}}+{\beta }_{2\mathit{K}}{X}_{\mathit{ft}}+{\beta }_{1\mathit{K}}{X}_{\mathit{rt}}}}{{\sum }_1^K{e}^{{\mathit{\beta }}_{1\mathit{K}-1}{X}_{\mathit{it}}+{\beta }_{2,\mathit{K}-1}{X}_{\mathit{ft}}+{\beta }_{3,\mathit{K}-1}{X}_{\mathit{rt}}}}$(5)

where $\mathit{K}$ is each of the four possible outcomes: (i) remain in the retail sector; (ii) switch to a different sector; (iii) become an entrepreneur; and (iv) become unemployed or become a student.

Individual ability, AKM fixed effects

Individual unobserved characteristics that are not included in the model could cause our estimates to be biased. When modelling longitudinal data, one would traditionally include an individual fixed effects component that absorbs time-invariant characteristics to avoid this bias. The drawback of using such a setup in our case is that we will not be able to assess the influence of different levels of human capital; instead, we will assess the influence of changes in human capital. To be able to analyse the influence of different levels of human capital while limiting the potential bias from unobserved heterogeneity, we use the method originally proposed by Abowd, Kramarz and Margolis (1999), referred to as the AKM Fixed Effects. In this method, individual and firm-level characteristics can be extracted by exploiting the information revealed by the change in wage when individuals change workplace, a.k.a. movers. It is also possible to extract information of those that do not change workplace, a.k.a. stayers, as long as they work in a firm that has had at least one mover. We estimate the following wage equation:

(6) $ln\left(\mathit{wag}{e}_{\mathit{it}}\right)={\alpha }_i+{\varphi }_{\mathit{it}}+\mathit{\beta }\mathit{X}+e_{\mathit{it}}$(6)

Where the dependent variable is the natural logarithm of the annual wage income for individual i at time t. ${\alpha }_i$ is the unobservable individual fixed effect and ${\varphi }_{\mathit{it}}$ is the firm fixed effects, and $\mathit{X}$ is a vector with all observable characteristics of the firm and individual (age, education, and year) and $e_{\mathit{it}}$ is a random error term. From this estimation, we recover the individual fixed effects, which we refer to as AKM-ability, and include this in the estimations. We only consider full-time employment. As the minimum wage negotiations only cover parts of the industries, we use the definition of low wage as an indicator for full-time jobs.¹² Summary statistics of the variables that are included in the estimation of equation (6)(6) $ln\left(\mathit{wag}{e}_{\mathit{it}}\right)={\alpha }_i+{\varphi }_{\mathit{it}}+\mathit{\beta }\mathit{X}+e_{\mathit{it}}$(6) can be found in the Appendix Table A2.

Empirical findings

The marginal effects for the logit and multinomial logit regressions are reported in Tables 1, 2 a and 2 b (the control variables are included in the estimations and the results with the full model are reported in the Appendix in Tables A3, A4, and A5). The logit estimations are separated into two columns, where the second column includes estimation with the AKM-ability variable.

Table 1. Marginal effects from the logit estimation. Column 2 presents the results when we control for AKM-ability variable.

Vars.	(1) Quit	(2) Quit
Tenure est.	−0.0155***	−0.0146***
	(0.000129)	(0.000127)
Tenure est. >16 years (0,1)	0.0912***	0.100***
	(0.0125)	(0.0122)
Tenure ind.	−0.00205***	−0.000523***
	(0.000128)	(0.000127)
Tenure ind. >16 years (0,1)	−0.0399***	−0.0516***
	(0.0127)	(0.0124)
Education Primary, 9 years^a	0.0384***	0.0401***
	(0.00192)	(0.00188)
Secondary,< 3 years	0.0559***	0.0586***
	(0.00179)	(0.00175)
Post-secondary, < 3 years	0.0713***	0.0741***
	(0.00196)	(0.00193)
Post-secondary, > 3 years	0.0916***	0.0884***
	(0.00211)	(0.00207)
Doctoral	0.142***	0.129***
	(0.0134)	(0.0131)
Retail education (0,1)	−0.0249***	−0.0233***
	(0.000857)	(0.000846)
Occup. skill: Level 2^b	−0.0489***	−0.0356***
	(0.00238)	(0.00228)
Level 3	−0.0865***	−0.0564***
	(0.00258)	(0.00254)
Level 4	−0.0981***	−0.0592***
	(0.00254)	(0.00257)
AKM-ability	N	Y
Year FE	Y	Y
Obs	2,122,736	2,087,858

Heteroscedasticity-robust standard errors, clustered on the establishment level, are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. Base categories are: ^aPrimary education, less than 9 years, ^bLevel 1.

Table 2a. Marginal effects from the multinomial logit estimation. Column 2 and 4 presents the results when we control for AKM-ability variable.

Vars.	(1) Stayed ind.	(2) Stayed ind.	(3) Left ind.	(4) Left ind.
Multinomial
Tenure est.	−0.0120***	−0.0121***	0.00738***	0.00720***
	(0.000358)	(0.000360)	(0.000338)	(0.000341)
Tenure est. >16 years (0,1)	0.00406	0.0247	0.00694	−0.0174
	(0.0412)	(0.0407)	(0.0356)	(0.0356)
Tenure ind.	0.0291***	0.0286***	−0.0270***	−0.0278***
	(0.000344)	(0.000351)	(0.000321)	(0.000327)
Tenure ind. >16 years (0,1)	−0.106**	−0.110***	0.0623*	0.0775**
	(0.0417)	(0.0412)	(0.0363)	(0.0362)
Education Primary 9 years^a	−0.0682***	−0.0713***	0.0800***	0.0799***
	(0.00631)	(0.00654)	(0.00549)	(0.00578)
Secondary,<3 years	−0.0789***	−0.0792***	0.107***	0.103***
	(0.00598)	(0.00621)	(0.00517)	(0.00546)
Post-secondary, <3 years	−0.0929***	−0.0857***	0.121***	0.110***
	(0.00629)	(0.00653)	(0.00549)	(0.00577)
Post-secondary >3 years	−0.138***	−0.134***	0.186***	0.178***
	(0.00654)	(0.00678)	(0.00577)	(0.00607)
Doctoral	−0.0871***	−0.0985***	0.127***	0.139***
	(0.0260)	(0.0270)	(0.0239)	(0.0255)
Retail ed. (0,1)	0.0172***	0.0179***	−0.0108***	−0.0125***
	(0.00232)	(0.00237)	(0.00216)	(0.00222)
occup. skill Level 2^b	0.0706***	0.0845***	−0.0284***	−0.0527***
	(0.00517)	(0.00551)	(0.00455)	(0.00506)
Level 3	0.0315***	0.0717***	0.0340***	−0.0229***
	(0.00591)	(0.00634)	(0.00531)	(0.00587)
Level 4	0.0516***	0.117***	−0.00164	−0.0846***
	(0.00583)	(0.00640)	(0.00518)	(0.00584)
AKM-ability	N	Y	N	Y
Year FE	Y	Y	Y	Y
Obs.	471,736	446,079	471,750	446,079

Heteroscedasticity-robust standard errors, clustered on the establishment level, are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. Base categories are: ^aPrimary education, less than 9 years, ^bLevel 1.

Table 2b. Marginal effects from the multinomial logit estimation. Column 6 and 8 presents the results when we control for AKM-ability variable.

Multinomial
Vars.	(5) Entrepreneur	(6) Entrepreneur	(7) Unemployed/ studying	(8) Unemployed/ studying
Tenure est.	0.00145***	0.00160***	0.00313***	0.00334***
	(6.58e-05)	(6.58e-05)	(0.000174)	(0.000163)
Tenure est. >16 years (0,1)	−0.00177	−0.00202	−0.00923	−0.00534
	(0.00393)	(0.00379)	(0.0157)	(0.0144)
Tenure ind.	−0.000843***	−0.000776***	−0.00128***	−7.82e-06
	(6.96e-05)	(7.01e-05)	(0.000175)	(0.000170)
Tenure ind. >16 years (0,1)	−0.00129	−0.00108	0.0453***	0.0338**
	(0.00406)	(0.00391)	(0.0159)	(0.0146)
Education Primary 9 years^a	0.00504***	0.00525***	−0.0168***	−0.0139***
	(0.000867)	(0.000891)	(0.00351)	(0.00343)
Secondary,<3 years	0.00469***	0.00466***	−0.0328***	−0.0284***
	(0.000789)	(0.000812)	(0.00338)	(0.00331)
Post-secondary, <3 years	0.00503***	0.00468***	−0.0329***	−0.0290***
	(0.000879)	(0.000902)	(0.00353)	(0.00346)
Post-secondary >3 years	0.00271***	0.00256***	−0.0506***	−0.0463***
	(0.000890)	(0.000914)	(0.00359)	(0.00351)
Doctoral	−0.000683	−0.00209	−0.0389***	−0.0384***
	(0.00258)	(0.00247)	(0.0127)	(0.0121)
Retail ed. (0,1)	−0.000267	−0.000431	−0.00614***	−0.00498***
	(0.000425)	(0.000424)	(0.00120)	(0.00116)
Occup. skill Level 2^b	0.00234***	0.00195**	−0.0446***	−0.0337***
	(0.000789)	(0.000845)	(0.00347)	(0.00337)
Level 3	0.0107***	0.00936***	−0.0762***	−0.0582***
	(0.000997)	(0.00106)	(0.00368)	(0.00363)
Level 4	0.0354***	0.0322***	−0.0853***	−0.0651***
	(0.00125)	(0.00143)	(0.00365)	(0.00366)
AKM-ability	N	Y	N	Y
Year FE	Y	Y	Y	Y
Obs.	471,736	446,079	471,736	446,079

Heteroscedasticity-robust standard errors, clustered on the establishment level, are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. Base categories are: ^aPrimary education, less than 9 years, ^bLevel 1.

Quitting

Table 1, column 1, presents the results from the regression analysis for the probability of quitting a job. Establishment tenure has a negative effect on the likelihood of quitting a job. The dummy for 16 years of experience or more has a positive effect, which indicates that the effect is negative but decreasing. This relationship is in line with the theory as well as with empirical studies on labor turnover in the retail sector, e.g. Gable, Hollon, and Dangello (1984); Schulz, Bigoness, and Gagnon (1987) but contradicts Hendrie (2004). Industry tenure also has a negative effect on the likelihood of quitting, which confirms the findings in previous studies such as Gable, Hollon, and Dangello (1984) and Min (2007). An increase in industry tenure with 1 year decreases the propensity to quit with 0.2%, while the corresponding effect of establishment tenure is 1.5%. Thus, the economic significance of industry tenure is relatively minor.

Compared with the baseline category of having less than nine years of primary education, the likelihood of quitting increases with the level of education. Secondary education of up to three years increases the probability of quitting by 4%, while having a doctoral degree increases the probability of quitting by 14%. This result is in line with the theory, where worker-specific human capital increases the expected value of leaving employment. If an individual has a retail education, the probability of quitting decreases by close to 2.5%. Thus, a specific education within retail appears to increase the willingness to stay in the current retail employment. This result is in line with the hypothesis that occupational investment, proxied by retail education, increases the opportunity cost to leave a firm. Compared to the lowest level of occupational skill, the likelihood of quitting is negative for all the higher occupational skills. In column 2, the AKM-ability variable is included. As this cannot be estimated for firms that have not had at least one mover, the sample size is now somewhat smaller. The results remain of a similar magnitude and direction as before.

The control variables show (results can be found in Appendix Table A3) that the likelihood of quitting a job decreases with age, and at an increasing rate, for males and for those who are foreign born, except individuals that originate from South America. Due to job discrimination (Grand and Szulkin 2002), foreign born people have lower labor mobility and may thus be less inclined to leave employment. The same reasoning also applies to aging, which tends to be negatively related to labor mobility. Our finding that males have a lower likelihood of quitting is surprising and goes against previous studies on employment turnover (Keith and McWilliams 1995; Loprest 1992).

The establishment-level relative wage as well as the regional relative wage have a negative relationship with the likelihood of quitting. If an individual has a wage that is relatively high to others in the same occupation in the same workplace, he or she is less likely to quit. A higher wage relative to others in the establishment and the region further indicates that the individual is better off to remain in the current job, thus he or she is less likely to quit, which is in line with the wealth maximization hypothesis. As before, the coefficients remain of the same magnitude and direction when including the AKM-ability variable.

Turnover destinations: paths after quitting

Tables 2a and 2b show the results on determinants of paths after quitting. Establishment tenure appears to decrease the likelihood of remaining in the retail sector and to increase the tendency to become employed in a different sector, while industry tenure has the opposite effect. Establishment tenure reflects firm-specific human capital, while industry tenure reflects industry-specific human capital, which is transferrable across firms. Consequently, industry tenure increases inter-firm and intra-industry mobility (Becker 1964). A similar analogy to that of firm-specific human capital may apply to industry-specific human capital such as industry experience. Thus, the industry-specific human capital is valued more in the same industry compared to others, which then reduces the mobility of individuals between industries. It is also in line with the argument of job embeddedness of Kiazad et al. (2015). Those with higher levels of education are more likely to remain in the retail sector, and less likely to become employed in another sector. The effect of education first increases as the level rises, and then it decreases again for doctoral level, which indicates a quadratic relationship, which is in line with findings in previous studies, e.g. Carless and Arnup (2011). These results are robust to the inclusion of AKM ability variable. Those with a retail education are more likely to remain in the retail sector and less likely to leave it. As changing sectors may implicate a change of occupation, changing sectors may be seen as an opportunity cost for those with a retail education, thus this result supports the findings of Zimmerman et al. (2020). With increasing levels of occupational skill, the likelihood of remaining in the retail sector also increases. This relationship holds also when including the AKM ability-variable. The effect of occupational skill is the opposite when considering the propensity to leave the sector. However, only the influence of one of the occupational skill levels, level two, is robust to the inclusion of the ability-variable.

Tenure in the previous workplace increases the probability of becoming an entrepreneur, which is in line with studies on antecedents of entrepreneurship (Mueller 2006). However, tenure in the industry has the opposite effect and decreases the propensity to go into self-employment. The economic significance of these relationships is rather minor; however, establishment tenure influences entrepreneurship with a mere 0.1% and industry tenure with 0.08%. Compared to the lowest level of education, all levels of education except doctoral, increase the probability to become an entrepreneur, which is opposite to the findings of Backman and Karlsson (2013). This effect increases in magnitude as the level rises, albeit the coefficients are of a small magnitude also here. Retail education has no effect on the probability of becoming an entrepreneur, while it decreases the probability of becoming unemployed or starting studying. When examining the occupational skill, the likelihood to become an entrepreneur increases with each level, and the opposite is found for the propensity to become unemployed or start studying. The results are robust to inclusion of the AKM ability variable.

Regarding the control variables (results can be found in Appendix Tables A4 and A5), when we examine wage-variables, the relative wage, within the establishment increases the probability to remain in the industry and to become an entrepreneur, while it decreases the probability of leaving the sector and becoming unemployed or start studying. In terms of demographic variables, we find that the probability to leave the sector, to become an entrepreneur and become unemployed or studying increases with age, while the probability to remain in the sector decreases. The effect of age on the probability of entrepreneurship is in line with previous studies, for instance, Backman and Karlsson (2013). Relative to males, females are more likely to remain in the retail sector, while they are less likely to switch to a different sector which is in line with studies showing that males are more mobile than females in the labor market, and are more prone to career changes than women (Blau and Lunz 1998; Carless and Arnup 2011). Compared to those who originate from Sweden, those born outside of Sweden are less likely to leave the sector (Keith and McWilliams 1995; Loprest 1992), more likely to remain in the retail sector, and to become unemployed or start studying. All the results in Tables A4 and A5 are robust to the inclusion of the AKM-ability variable.

Conclusion

Our findings show the importance of vocational training programs targeting different sectors. Thus, occupational investment tends to be important. If these types of programs and other occupational investments were made more attractive for students it could help to reduce labor turnover in retail firms and to retain workers in the retail sector. It is of course important that these vocational training programs are designed in close collaboration with the industry to equip students with the right skills, expectations, and knowledge about the sector. This is especially important given the many challenges that retail is facing. Many countries, Sweden included, are experiencing a retail sector that is facing shrinking margins, fierce competition, and a fast-changing landscape where online retail is something that many firms must adapt to. The geographical pattern has also changed over time, where city-center locations have been more challenged by out-of-town areas (Forsberg, 1998; Kärrholm, 2016) leading to a high street decline. The sector is known for a tight labour market, and concerns about low-skilled employment are widely recognised. As such, the role of human capital is likely to grow in importance over time. As the skills needed in the retail sector become more advanced, and there is room for governmental investments to support and upgrade these programs.

As individuals with more complex working tasks are less likely to quit their jobs and less likely to leave the retail sector, we argue that firms should invest in internal training programs to enable employees to advance their competence and to retain their knowledge at the workplace and in the retail sector. The results further point to the importance of demographics and relative wage for the further career development of individuals in this sector. Our findings also have practical relevance for recruiting managers within the retail sector, hiring individuals with vocational training and more industry experience lowers the probability that the newly hired would leave the firm.

How applicable are our findings to other countries? The Swedish labor market is characterized by a regulatory system that gives workers exceptionally strong rights. Thus, terminating employment from the employers’ point of view is relatively difficult. Moreover, by excluding workers in plants that closed and examining only those that worked full-time in the sector, we have reasons to believe that the majority of the job separations were voluntary. This means that our results can be expected to be more pronounced in countries where labor markets have less extensive workers’ rights, and the results are thus applicable to many other developed countries. Like most other studies, ours is not without caveats that could be used to inform future studies about possible research avenues. In this study, we exclude part-time workers. As noted earlier, they are an important part of the employment in retail and retention of such employees is thus important and should be investigated in future research. Although our measures of human capital capture detailed information about the employees there is still room to expand our knowledge on how different types of human capital relate to the probability to quit a retail employment and turnover destination. Interesting aspects could be to further explore different vocational educations related to retail but also try to assess other vocational educations and degrees that influence a retail employee and his/her career trajectories. Another area for future research would be to study the mechanisms on a disaggregated industry level. As retail consists of several subsectors with a large degree of heterogeneity, e.g. grocery-, fashion- and electronics, the role of human capital may vary across these sectors. Thus, future research could explore the role of human capital in turnover destinations across such subsectors. Lastly, in the analysis we control for ascribed characteristics of the employees; however, the heterogeneity across employees can be further explored by conducting studies on different sets of retail employee populations such as: men/women or high/low educated.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The work was supported by the Hakon Swenson Stiftelsen and The Swedish Retail and Wholesale Council (Handelsrådet). Career paths in the wholesale trade. Who starts and ends – as employees and entrepreneurs?

Notes on contributors

Helena Nilsson

Helena Nilsson is an Assistant Professor of Economics at Jönköping International Business School, Jönköping, Sweden. She is also affiliated with the Centre for Entrepreneurship and Spatial Economics (CEnSE) in Jönköping and the Institute of Retail Research in Stockholm. Her research interests involve the spatial organization of consumer goods and services. She earned her bachelor’s and master’s degrees in Economics at the School of Business, Economics and Law at the University of Gothenburg, and her doctoral degree in Economics at Jönköping International Business School, Jönköping, Sweden.

Mikaela Backman

Mikaela Backman is an Associate Professor of Economics at Jönköping International Business School, Jönköping, Sweden. She is also affiliated with the Centre for Entrepreneurship and Spatial Economics (CEnSE) in Jönköping. Her research interests involve entrepreneurship, human capital, and regional economics. She earned her bachelor’s, master’s, and doctoral degrees in Economics at Jönköping International Business School.

Notes

1. Data for 2015, from http://handelsradet.se/wp-content/uploads/2016/01/Snabbfakta-2015.pdf. and http://www.statistikdatabasen.scb.se/pxweb/en/ssd/START__AM__AM0401__AM0401A/NAKUBefolkning2Ar/table/tableViewLayout1/,Accessed on 2019-11-11.

2. https://handelsradet.se/wp-content/uploads/2017/06/Handelns-betydelse-da-nu-och-i-framtiden_HER-rapport-2017.pdf, Accessed 2020-10-22.

3. https://handelssite.wordpress.com/2017/11/29/okad-delaktighet-minskar-personalomsattning/, Accessed 2020-10-22.

4. This is captured at the establishment level. However, we use the term ‘firm-specific’ to follow the concepts used in the human capital literature.

5. Occupational complexity refers to the complexity of work tasks that an occupation typically involves.

6. Establishments in the retail sector have an SIC between 47,111 and 47,792, which is constructed by Statistics Sweden.

7. Retrieved from https://arbetet.se/2006/10/27/tjnstemannafack-har-smsta-lgstalnerna/. Accessed 2020-11-12.

8. Values are adjusted for inflation annually.

9. Comparable data on country level is only available on trade sector level (retail and wholesale).

10. Booth and Hamer (2007) do, however,not find firm size is related to labor turnover.

11. The results are robust to estimation with linear probability model (OLS) (available upon request).

12. In 2018 below 0.1% of employed in Sweden had a full-time equivalent wage corresponding to 50% of the national median wage (Swedish National Mediation Office 2019).

Appendix A

Table A1. Descriptive statistics.

Variables	Obs.	Mean	Std. Dev.	Min.	Max.
Dependent variables
Quit (0,1)	2,461,047	0.226	0.418	0	1
Quit, next (0,1)
Stayed ind. (0,1)	556,862	0.569	0.495	0	1
Left ind. (0,1)	556,862	0.299	0.458	0	1
Entrepreneur (0,1)	556,862	0.041	0.198	0	1
Unemployed/studying (0,1)	556,862	0.091	0.288	0	1
Individual characteristics
Human capital
Est. tenure	2,655,398	4.846	3.893	1	17
Est. tenure >16 years (0,1)	2,655,398	0.035	0.183	0	1
Ind. tenure	2,654,900	6.368	4.246	1	17
Ind. tenure >16 years (0,1)	2,654,900	0.033	0.180	0	1
Education
1: Primary, < 9 years	2,646,673	0.027	0.163	0	1
2: Primary, 9 years	2,646,673	0.115	0.319	0	1
3: Secondary, ≤ 3 years	2,646,673	0.651	0.477	0	1
4: Post-secondary, ≤ 3 years	2,646,673	0.117	0.321	0	1
5: Post-secondary, ≥ 3 years	2,646,673	0.089	0.285	0	1
6: Doctoral	2,646,673	0.001	0.030	0	1
Retail education (0,1).	2,646,673	0.185	0.388	0	1
Control variables
Occup. skill
Level 1	2,470,779	0.023	0.150	0	1
Level 2	2,470,779	0.740	0.439	0	1
Level 3	2,470,779	0.091	0.288	0	1
Level 4	2,470,779	0.146	0.353	0	1
Relative earnings, est.	2,471,009	0.077	0.205	−2.868	2.428
Relative earnings, LA-region	2,471,009	0.156	0.265	−2.178	3.359
Married (0,1)	2,655,398	0.532	0.499	0	1
Children (0,1)	2,655,397	0.366	0.482	0	1
Age cat.
<25 years	2,655,398	0.137	0.344	0	1
≥25 & ≤40 years	2,655,398	0.402	0.490	0	1
>40 years	2,655,398	0.461	0.498	0	1
Female (0,1)	2,655,398	0.588	0.492	0	1
Region of birth
Swedish	2,655,283	0.890	0.313	0	1
North Am. Eur.	2,655,283	0.057	0.231	0	1
Africa	2,655,283	0.005	0.071	0	1
South America	2,655,283	0.006	0.080	0	1
Asia	2,655,283	0.041	0.198	0	1
Oceania	2,655,283	0.0003	0.018	0	1
Chain	2,655,398	0.537	0.499	0	1
Marginal return	2,472,750	−0.129	22.363	−8870.25	803
Regional characteristics
Retail region	2,655,398	29937.35	29213.38	37	84437
Centrality
Category 1: Rural: very remote and sparsely populated	2,655,398	0.008	0.087	0	1
Category 2: Rural: remote and sparsely populated	2,655,398	0.049	0.215	0	1
Category 3: Rural: close to a larger city	2,655,398	0.069	0.253	0	1
Category 4: Urban: remote	2,655,398	0.068	0.251	0	1
Category 5: Urban: close to a larger city	2,655,398	0.435	0.496	0	1
Category 6: Metropolitan	2,655,398	0.372	0.483	0	1

Table A2. Summary statistics for variables included in the AKM estimation, equation 6.

Variables	Obs.	Mean	Std. Dev.	Min.	Max.
ln(Wage)	4,794,924	7.970	0.355	7.293	13.2424
Education
Primary, < 9 years	4,780,732	0.021	0.142	0	1
Primary, 9 years	4,780,732	0.104	0.306	0	1
Secondary, ≤ 3 years	4,780,732	0.624	0.484	0	1
Post-secondary, ≤ 3 years	4,780,732	0.129	0.335	0	1
Post-secondary, ≥ 3 years	4,780,732	0.120	0.325	0	1
Doctoral	4,780,732	0.001	0.037	0	1
Occup. skill
Level 1	4,430,767	0.037	0.188	0	1
Level 2	4,430,767	0.650	0.477	0	1
Level 3	4,430,767	0.141	0.348	0	1
Level 4	4,430,767	0.173	0.378	0	1
Married (0,1)	4,794,924	0.533	0.499	0	1
Children (0,1)	4,794,921	0.394	0.489	0	1
Age category:
< 25 years	4,794,924	0.123	0.328	0	1
>=25 & <= 40 years	4,794,924	0.441	0.497	0	1
> 40 years	4,794,924	0.436	0.496	0	1
Est. size	4,686,460	136.697	625.436	1	19231
Chain	4,686,460	0.517	0.500	0	1
Marginal return	4,060,588	−0.261	69.737	−49078.500	14140
Centrality
Category 1	4,794,924	0.008	0.088	0	1
Category 2	4,794,924	0.047	0.211	0	1
Category 3	4,794,924	0.067	0.251	0	1
Category 4	4,794,924	0.064	0.245	0	1
Category 5	4,794,924	0.415	0.493	0	1

Table A3. Marginal effects from the logit estimation estimations with control variables. Column 2 presents the results when we control for AKM-ability variable.

Vars.	(1) Quit	(2) Quit
Tenure est.	−0.0155***	−0.0146***
	(0.000129)	(0.000127)
Tenure est. > 16 years (0,1)	0.0912***	0.100***
	(0.0125)	(0.0122)
Tenure ind.	−0.00205***	−0.000523***
	(0.000128)	(0.000127)
Tenure ind. > 16 years (0,1)	−0.0399***	−0.0516***
	(0.0127)	(0.0124)
Education Primary, 9 years^a	0.0384***	0.0401***
	(0.00192)	(0.00188)
Secondary,< 3 years	0.0559***	0.0586***
	(0.00179)	(0.00175)
Post-secondary, < 3 years	0.0713***	0.0741***
	(0.00196)	(0.00193)
Post-secondary, > 3 years	0.0916***	0.0884***
	(0.00211)	(0.00207)
Doctoral	0.142***	0.129***
	(0.0134)	(0.0131)
Retail education (0,1)	−0.0249***	−0.0233***
	(0.000857)	(0.000846)
Occup. skill: Level 2 ^b	−0.0489***	−0.0356***
	(0.00238)	(0.00228)
Level 3	−0.0865***	−0.0564***
	(0.00258)	(0.00254)
Level 4	−0.0981***	−0.0592***
	(0.00254)	(0.00257)
Relative wage, est.	−0.0313***	−0.0215***
	(0.00183)	(0.00182)
Relative wage, LA-region	−0.135***	−0.0950***
	(0.00161)	(0.00176)
Female (0,1)	0.0303***	0.0270***
	(0.000617)	(0.000612)
Region of birth North Am. Eur. ^d	−0.0139***	−0.0150***
	(0.00131)	(0.00129)
Africa	−0.00682	−0.0124***
	(0.00423)	(0.00414)
South America	0.0117***	0.00818**
	(0.00370)	(0.00363)
Asia	0.00106	−0.00507***
	(0.00155)	(0.00152)
Oceania	0.0190	0.0144
	(0.0173)	(0.0168)
Age cat. >= 25 & <= 40 years^c	−0.0409***	−0.0322***
	(0.00109)	(0.00108)
> 40 years	−0.132***	−0.123***
	(0.00114)	(0.00113)
Married (0,1)	−0.00401***	−0.00351***
	(0.000730)	(0.000726)
Children (0,1)	0.0101***	0.0125***
	(0.000756)	(0.000750)
Chain (0,1)	0.0312***	0.0322***
	(0.000612)	(0.000606)
Size of est.	−8.31e-05***	−7.73e-05***
	(2.43e-06)	(2.39e-06)
Marginal return	2.90e-05**	2.51e-05*
	(1.38e-05)	(1.29e-05)
Retail region	2.85e-07***	2.98e-07***
	(1.51e-08)	(1.49e-08)
Centrality, Category 2^e	−0.00585	−0.00419
	(0.00378)	(0.00371)
Category 3	−0.0150***	−0.0123***
	(0.00372)	(0.00366)
Category 4	−0.00143	0.00100
	(0.00371)	(0.00364)
Category 5	0.00178	0.00442
	(0.00358)	(0.00352)
Category 6	0.0267***	0.0292***
	(0.00369)	(0.00362)
AKM-ability	N	Y
Year FE	Y	Y
Obs	2,122,736	2,087,858

Heteroscedasticity-robust standard errors, clustered on the establishment level, are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. Base categories are: ^aPrimary education, less than 9 years, ^bLevel 1, ^cAge category: <25; ^dSweden. ^eRural, remote municipality.

Table A4. Marginal effects from the multinomial logit estimation. Columns 2 and 4 present the results when we control for AKM-ability variable.

Vars.	(1) Stayed ind.	(2) Stayed ind.	(3) Left ind.	(4) Left ind.
Multinomial
Tenure est.	−0.0120***	−0.0121***	0.00738***	0.00720***
	(0.000358)	(0.000360)	(0.000338)	(0.000341)
Tenure est. >16 years(0,1)	0.00406	0.0247	0.00694	−0.0174
	(0.0412)	(0.0407)	(0.0356)	(0.0356)
Tenure ind.	0.0291***	0.0286***	−0.0270***	−0.0278***
	(0.000344)	(0.000351)	(0.000321)	(0.000327)
Tenure ind. >16 years (0,1)	−0.106**	−0.110***	0.0623*	0.0775**
	(0.0417)	(0.0412)	(0.0363)	(0.0362)
Education Primary 9 years^a	−0.0682***	−0.0713***	0.0800***	0.0799***
	(0.00631)	(0.00654)	(0.00549)	(0.00578)
Secondary,<3 years	−0.0789***	−0.0792***	0.107***	0.103***
	(0.00598)	(0.00621)	(0.00517)	(0.00546)
Post-secondary, <3 years	−0.0929***	−0.0857***	0.121***	0.110***
	(0.00629)	(0.00653)	(0.00549)	(0.00577)
Post-secondary >3 years	−0.138***	−0.134***	0.186***	0.178***
	(0.00654)	(0.00678)	(0.00577)	(0.00607)
Doctoral	−0.0871***	−0.0985***	0.127***	0.139***
	(0.0260)	(0.0270)	(0.0239)	(0.0255)
Retail ed. (0,1)	0.0172***	0.0179***	−0.0108***	−0.0125***
	(0.00232)	(0.00237)	(0.00216)	(0.00222)
Occup. skill Level 2^b	0.0706***	0.0845***	−0.0284***	−0.0527***
	(0.00517)	(0.00551)	(0.00455)	(0.00506)
Level 3	0.0315***	0.0717***	0.0340***	−0.0229***
	(0.00591)	(0.00634)	(0.00531)	(0.00587)
Level 4	0.0516***	0.117***	−0.00164	−0.0846***
	(0.00583)	(0.00640)	(0.00518)	(0.00584)
Relative wage, est.	0.103***	0.114***	−0.0666***	−0.0802***
	(0.00460)	(0.00476)	(0.00426)	(0.00442)
Relative wage, LA region	−0.0570***	0.00945**	0.166***	0.0818***
	(0.00398)	(0.00455)	(0.00368)	(0.00421)
Female (0,1)	0.181***	0.185***	−0.142***	−0.148***
	(0.00165)	(0.00170)	(0.00156)	(0.00161)
Region of birth North Am. Eur.^d	0.0251***	0.0249***	−0.0464***	−0.0421***
	(0.00352)	(0.00360)	(0.00313)	(0.00324)
Africa	0.0455***	0.0458***	−0.0837***	−0.0769***
	(0.0106)	(0.0109)	(0.00868)	(0.00917)
South America	0.0163*	0.0121	−0.0342***	−0.0278***
	(0.00863)	(0.00891)	(0.00770)	(0.00804)
Asia	0.0407***	0.0468***	−0.0921***	−0.0858***
	(0.00383)	(0.00395)	(0.00313)	(0.00330)
Oceania	−0.0621	−0.0557	−0.0696**	−0.0433
	(0.0412)	(0.0423)	(0.0344)	(0.0373)
>= 25 & <= 40 years^c	−0.0518***	−0.0593***	0.0328***	0.0366***
	(0.00203)	(0.00209)	(0.00187)	(0.00193)
> 40 years	−0.153***	−0.157***	0.0718***	0.0766***
	(0.00256)	(0.00263)	(0.00236)	(0.00243)
Married (0,1)	0.0308***	0.0364***	−0.0382***	−0.0420***
	(0.00192)	(0.00199)	(0.00179)	(0.00186)
Children (0,1)	−1.13e-05	0.00366*	0.0419***	0.0368***
	(0.00200)	(0.00206)	(0.00186)	(0.00193)
Chain (0,1)	0.139***	0.142***	−0.0954***	−0.101***
	(0.00161)	(0.00165)	(0.00148)	(0.00153)
Size of est.	−0.000182***	−0.000178***	0.000229***	0.000224***
	(7.22e-06)	(7.26e-06)	(5.97e-06)	(6.06e-06)
Marginal return	0.00105	0.00124	−0.000857	−0.000998
	(0.000930)	(0.00136)	(0.000695)	(0.00103)
Retail region	−5.56e-08	−7.57e-08*	2.17e-07***	2.30e-07***
	(3.96e-08)	(4.08e-08)	(3.64e-08)	(3.76e-08)
Centrality, Category 2^e	0.0468***	0.0411***	−0.0497***	−0.0431***
	(0.0107)	(0.0112)	(0.0103)	(0.0107)
Category 3	0.0605***	0.0527***	−0.0608***	−0.0499***
	(0.0106)	(0.0110)	(0.0102)	(0.0106)
Category 4	0.0394***	0.0299***	−0.0416***	−0.0307***
	(0.0105)	(0.0109)	(0.0101)	(0.0105)
Category 5	0.0575***	0.0469***	−0.0606***	−0.0482***
	(0.0101)	(0.0105)	(0.00974)	(0.0101)
Category 6	0.0482***	0.0373***	−0.0660***	−0.0521***
	(0.0104)	(0.0108)	(0.00997)	(0.0104)
AKM-ability	N	Y	N	Y
Year FE	Y	Y	Y	Y
Obs.	471,736	446,079	471,750	446,079

Heteroscedasticity-robust standard errors, clustered on the establishment level, are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. Base categories are: ^aPrimary education, less than 9 years, ^bLevel 1, ^cAge category: < 25; ^dSweden. ^eRural, remote municipality

Table A5. Marginal effects from the multinomial logit estimation. Columns 2 and 4 present the results when we control for AKM-ability variable.

Multinomial
Vars.	(1) Entrepreneur	(2) Entrepreneur	(3) Unemployed/ studying	(4) Unemployed/ studying
Tenure est.	0.00145***	0.00160***	0.00313***	0.00334***
	(6.58e-05)	(6.58e-05)	(0.000174)	(0.000163)
Tenure est. >16 years(0,1)	−0.00177	−0.00202	−0.00923	−0.00534
	(0.00393)	(0.00379)	(0.0157)	(0.0144)
Tenure ind.	−0.000843***	−0.000776***	−0.00128***	−7.82e-06
	(6.96e-05)	(7.01e-05)	(0.000175)	(0.000170)
Tenure ind. >16 years (0,1)	−0.00129	−0.00108	0.0453***	0.0338**
	(0.00406)	(0.00391)	(0.0159)	(0.0146)
Education Primary 9 years^a	0.00504***	0.00525***	−0.0168***	−0.0139***
	(0.000867)	(0.000891)	(0.00351)	(0.00343)
Secondary,<3 years	0.00469***	0.00466***	−0.0328***	−0.0284***
	(0.000789)	(0.000812)	(0.00338)	(0.00331)
Post-secondary, <3 years	0.00503***	0.00468***	−0.0329***	−0.0290***
	(0.000879)	(0.000902)	(0.00353)	(0.00346)
Post-secondary >3 years	0.00271***	0.00256***	−0.0506***	−0.0463***
	(0.000890)	(0.000914)	(0.00359)	(0.00351)
Doctoral	−0.000683	−0.00209	−0.0389***	−0.0384***
	(0.00258)	(0.00247)	(0.0127)	(0.0121)
Retail ed. (0,1)	−0.000267	−0.000431	−0.00614***	−0.00498***
	(0.000425)	(0.000424)	(0.00120)	(0.00116)
Occup. skill Level 2^b	0.00234***	0.00195**	−0.0446***	−0.0337***
	(0.000789)	(0.000845)	(0.00347)	(0.00337)
Level 3	0.0107***	0.00936***	−0.0762***	−0.0582***
	(0.000997)	(0.00106)	(0.00368)	(0.00363)
Level 4	0.0354***	0.0322***	−0.0853***	−0.0651***
	(0.00125)	(0.00143)	(0.00365)	(0.00366)
Relative wage, est.	0.00793***	0.00749***	−0.0443***	−0.0414***
	(0.000859)	(0.000866)	(0.00247)	(0.00243)
Relative wage, LA-region	−0.00478***	−0.00572***	−0.104***	−0.0855***
	(0.000648)	(0.000727)	(0.00220)	(0.00238)
Female (0,1)	−0.0136***	−0.0132***	−0.0248***	−0.0241***
	(0.000422)	(0.000427)	(0.000874)	(0.000859)
Region of birth North Am. Eur. ^d	−0.00195***	−0.00213***	0.0232***	0.0193***
	(0.000616)	(0.000622)	(0.00205)	(0.00200)
Africa	−0.00443**	−0.00423**	0.0426***	0.0354***
	(0.00186)	(0.00198)	(0.00660)	(0.00657)
South America	−0.00491***	−0.00464***	0.0229***	0.0204***
	(0.00155)	(0.00159)	(0.00499)	(0.00488)
Asia	0.00386***	0.00382***	0.0475***	0.0351***
	(0.000861)	(0.000907)	(0.00250)	(0.00244)
Oceania	0.0113	0.0146	0.120***	0.0844***
	(0.0112)	(0.0124)	(0.0337)	(0.0317)
Age cat. >= 25 & <= 40 years^c	0.0118***	0.0115***	0.00718***	0.0111***
	(0.000354)	(0.000355)	(0.000945)	(0.000919)
> 40 years	0.0303***	0.0287***	0.0513***	0.0519***
	(0.000687)	(0.000676)	(0.00136)	(0.00133)
Married (0,1)	0.00533***	0.00489***	0.00212**	0.000727
	(0.000361)	(0.000366)	(0.000973)	(0.000953)
Children (0,1)	0.000562*	0.000381	−0.0424***	−0.0409***
	(0.000342)	(0.000345)	(0.00107)	(0.00104)
Chain (0,1)	−0.0287***	−0.0277***	−0.0146***	−0.0132***
	(0.000413)	(0.000415)	(0.000824)	(0.000810)
Size of est.	−6.66e-05***	−6.26e-05***	1.95e-05***	1.69e-05***
	(5.13e-06)	(4.95e-06)	(3.39e-06)	(3.29e-06)
Marginal return	−4.38e-05	−5.28e-05	−0.000152	−0.000191
	(4.25e-05)	(6.02e-05)	(0.000199)	(0.000270)
Retail region	3.88e-08***	3.43e-08***	−2.00e-07***	−1.89e-07***
	(7.55e-09)	(7.66e-09)	(2.10e-08)	(2.08e-08)
Centrality, Category 2^e	−0.00231	−0.00262	0.00518	0.00463
	(0.00208)	(0.00213)	(0.00442)	(0.00435)
Category 3	−0.00286	−0.00355*	0.00313	0.000739
	(0.00204)	(0.00209)	(0.00435)	(0.00427)
Category 4	−0.00580***	−0.00622***	0.00802*	0.00701
	(0.00203)	(0.00209)	(0.00433)	(0.00426)
Category 5	−0.00638***	−0.00667***	0.00943**	0.00804**
	(0.00197)	(0.00203)	(0.00413)	(0.00407)
Category 6	−0.00779***	−0.00801***	0.0256***	0.0228***
	(0.00203)	(0.00208)	(0.00429)	(0.00423)
AKM-ability	N	Y	N	Y
Year FE	Y	Y	Y	Y
Obs.	471,736	446,079	471,736	446,079

Heteroscedasticity-robust standard errors, clustered on the establishment level, are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. Base categories are: ^aPrimary education, less than 9 years, ^bLevel 1, ^cAge category: <25; ^dSweden. ^eRural, remote municipality.