2.1 Perceived overqualification and employee performance

The concept of overqualification can be examined in two different ways: subjectively and objectively. From the objective perspective, overqualification refers to the standard of skills that individuals must have and an educational level of the individual that is higher than that required by the work (^{Fine, 2007}; ^{Maltarich, Reilly, & Nyberg, 2011}). From the subjective perspective, on the other hand, overqualification is the perception that the skills of an individual exceed what is required by the work (Maltarich et al., 2011). Subjective overqualification refers to individuals' perceptions of whether they are overqualified for their current jobs or not (^{Erdogan et al., 2011}). Individuals’ perceptions of their own overqualification may not always reflect the reality. In other words, an individual who does not have the knowledge, skills, abilities, and education level required for the work may consider himself to have those qualifications. This comes to the forefront as the main difference that distinguishes subjective overqualification from objective overqualification (Maltarich et al., 2011:236-237; ^{Maynard et al., 2015}:211).

The concept of perceived overqualification was discussed by ^{Kristof-Brown (2000}:53) in the context of demand-ability mismatch. The term “demand” represents the requirements of the work within the organization, while “talent” refers to the level of knowledge, skills, training, experience, and other possible qualifications of the person performing the work. In other words, demand-ability mismatch means that the individual has more knowledge, skills, training, experience, and other possible qualifications related to the work. Therefore, the current study focuses on the concept of perceived overqualification. Person-job (PJ) cohesion is one of the narrower scopes of person-environment cohesion (PE) and expresses the harmony between an employee's characteristics and the job or task (Krıstof-Brown, Zımmerman, and Johnson, 2005). There are basically two types of PJ compatibility. One of these is demands-talents, and the other is needs-facts (Krıstof-Brown et al., 2005; Arvan, Pindekb, Andelc, & Spectora, 2019). What is addressed here is the demand-ability fit. In the context of perceived overqualification, there is no person-job fit. As a result, it is observed that individuals experience lower organizational commitment and job satisfaction (^{Fine and Nevo, 2008}:346). Furthermore, it is regarded as a perception that escalates the individual's turnover intentions (^{Maynard and Parfyovona, 2013}:437; ^{Erdogan & Bauer, 2009}:558; Maynard et al., 2006:549).

Performance is a concept that defines the extent to which individuals can make use of their potential or actual knowledge and skills to achieve their goals or expectations. In other words, it is the percentage of a person’s use of his/her capacity to successfully complete a task within a certain period of time (^{Kaynak et al., 1998}). More specifically, the concept of performance is defined as the degree to which the goal is achieved. Performance refers to what extent an organizational goal is reached by an individual, group, work unit, or firm through performing predefined tasks (^{Kasnaklı, 2002}).

^{Çoban and Eren (2017}) examined employee performance under the framework of organizational performance. According to Kuebg and Krahn (1999), ^{Kueng (2000}), Eren (2007), ^{Zerenler and İraz (2006}), and Özdemir and Yılmaz (2016), employee performance is one of the factors that increase or decrease organizational performance. In other words, performance demonstrated by employees plays a great role in paving the way for an organization to reach its goals. Employee performance, which is of great importance for businesses, can be improved through a number of activities. Compulsory practical training provided in culinary schools and universities plays an effective role in increasing the performance of chefs, which have been selected as the sample group for the current study (^{Bucak & Yiğit, 2018}). Furthermore, on-the-job training, motivation-boosting meetings, increases in salary, and promotions are the sorts of factors that increase employee performance (Yılmaz, 2018:103; ^{Uygur, 2007}).

In the literature, it is observed that person-job fit is associated with job performance (^{Kristof-Brown et al., 2005}:317). The relationship between perceived overqualification and employee performance was discussed in the study of ^{Ünsal Akbıyık (2016}a), and it was concluded that perceived overqualification had a negative effect on employee performance. From this point of view, it is predicted in the current study that perceived overqualification is associated with employee performance, and in this context the following hypothesis is proposed to reveal the relationship between the concepts in question.

Hypothesis 1: Perceived overqualification has a negative effect on chefs’ performance.

As indicated in the conceptual model of the current study (see Figure 1), it is predicted that perceived overqualification will have an impact on employee performance.

Figure 1 Conceptual model 

3.1. Research variables and measurement

In order to measure the concept of perceived overqualification, which is the independent variable in the current study, a scale developed by ^{Maynard et al. (2006}) is used. The scale consists of nine items that assess perceived overqualification, as follows: “My job requires less education than I have”; “The work experience that I have is not necessary to be successful on this job”; “I have job skills that are not required for this job”; “Someone with less education than myself could perform well on my job”; “My previous training is not being fully utilized on this job”; “I have a lot of knowledge that I do not need in order to do my job”; “My education level is above the education level required by my job”; “Someone with less work experience than myself could do my job just as well”; and “I have more abilities than I need in order to do my job.” The original scale is in the format of a 7-point Likert scale. Assessments range from “1 - strongly disagree” to “7 - strongly agree.”

The dependent variable in the current study is employee performance. In order to measure the performance of kitchen staff in hotel businesses, the employee performance scale developed by ^{Kirkman and Rosen (1999}:67) and adapted by ^{Sigler and Pearson (2000}:34) is used. The adapted scale consists of five items, whereas the original scale is in the format of a 7-point Likert scale. Employee performance was assessed using the 5-item scale of employee performance, as follows: “I complete my tasks on time”; “I meet my goals”; “I exceed my goals”; “I make sure that products meet or exceed quality standards”; and “I respond quickly when problems come up.” Assessments range from “1 - strongly disagree” to “7 - strongly agree.”

As a result of the comprehensive literature review on the scales, it was observed that a great number of studies tested and verified the reliability and validity of the scales used to measure perceived overqualification and employee performance. For instance, the performance scale was used by ^{Kirkman and Rosen (1999}), ^{Sigler and Pearson (2000}), and Çöl (2008), while the perceived overqualification scale was used by ^{Ünsal Akbıyık (2016}); ^{Alfes, Shantz, and Van Baalen (2016}); and ^{Liu, Luksyte, Zhou, Shi, & Wang (2015}). Therefore, it was decided to adopt the scales in question in the current study as well.

3.2 Sample and data collection

In the current study, the convenience sampling method, which is a non-probability sampling technique, was used as sampling tool. In this sampling method, the researcher collects data from the most easily accessible subjects until reaching the sample size required for the research (^{Gürbüz & Şahin, 2015}:130). The reason underlying the preference of this sampling method in the current study is to conduct the questionnaire with as many eligible chefs as possible to reach in a short time the sample size that will represent the research population. In this sampling technique, the ability to represent the research population is low. Since the data collected from the same population in different time periods may differ, the collected data cannot be generalized over the population (^{İslamoglu, 2009}).

The research population consists of chefs working in 4- and 5-star hotels in Nevşehir. According to the data dated October 30, 2018, obtained from the Nevşehir Provincial Directorate of Culture and Tourism, there are 17 4-star and six 5-star hotels operating in Nevşehir. The research was conducted throughout January and February of 2019. A total of 200 questionnaires was handed out to kitchen chefs and managers at hotels. Of these, 153 questionnaires were returned. However, it was observed that 15 questionnaires were left blank or never filled in. Therefore, those forms were excluded from the sample. The number of eligible questionnaires was 138. In order to use structural equation modeling, the minimum number of samples should be 5 to 10 times greater than the number of observed variables (^{Gürbüz & Şahin, 2015}:324; ^{Reisinger & Mavondo, 2007}:42; ^{Hair, Anderson, Babin, & Black, 2010}). The number of observed variables in this study is 14. As the number of observed variables is multiplied by 10 (14 * 10 = 140), the result indicates that the structural equation modeling can be performed in the current study.

3.3. Data analysis

The data collected for this study were analyzed in a computer environment using SPSS and AMOS programs. In the study, normal distribution of data was tested via scatter plot matrix and also on the basis of the skewness and kurtosis values. Since the research data showed normal distribution, it was continued with parametric analyses. The scales adopted in the current study were developed and used in previous studies; therefore, confirmatory factor analysis was performed to test the reliability of the scales. As a result of confirmatory factor analysis, two items violating the reliability were excluded from the analytical process. Following the exclusion of the items, the normal distribution of the scales was re-tested, and reliability analysis was re-performed. Finally, structural equation modeling (SEM) was conducted since it allows error variances to be included in the model and its explanatory power is statistically greater than regression analysis.

4.1. Descriptive statistics of respondent employees

Table 1 provides the descriptive statistical data (frequency and percentage [%]) of the employees who form the sample in the current study.

Table 1 Descriptive statistics of the respondents 

Regarding the gender of the employees involved in the study, it was observed that the majority (87) were men, who represented 63% of overall respondents. It was noted that the number of women respondents in the study was 50 with a rate of 36.2%. It was observed that 0.7% of the employees who participated in the study left this item unmarked. With regard to the age of the respondent employees, it was observed that 44.9% of overall respondents were aged between 21 and 30 years. In other words, the majority of the respondent employees in the current study was composed of young chefs. Upon examining the other statistics, it was noted that three respondent employees were aged 20 years or below (2.2%), 47 employees were aged 31-40 (34.1%), 20 employees were aged 41-50, and two employees were aged 50 or above (1.4%). It was observed that four (2.9%) respondent employees left this item unmarked. Regarding the educational status of the respondent employees, it was observed that the majority (62 respondents) was composed of chefs with bachelor’s and associate degrees. Upon examining the other statistics, it was noted that the number of respondents with primary and high school education was 47 (34.1%), and the number of those with master’s and doctoral degrees was 21 (15.2%). The remaining eight respondents left this item unmarked. Regarding the enterprise status of the hotels where respondent employees worked, the majority of respondents (86 respondents) were employed as chefs at 4-star hotels. The number of respondents from the kitchen departments of 5-star hotels was 48 (31.2%). The remaining four respondents (2.9%) left this item unmarked. Regarding the respondents’ years of professional experience, it was observed that the majority (43 respondents) had 6-10 years of experience in the sector. Other respondents’ years of experience are as follows: 35 respondents had less than five years (25.4%), 39 respondents had 11-15 years, 13 respondents had 16-20 years (9.4%), and six respondents had 21 years or more (1.4%). It was observed that the majority of employees (110 respondents) (79.7%) had training backgrounds in the field of tourism. On the other hand, it was noted that 28 respondents (20.3%) had no training experience in the field. As the respondents were asked about the number of tourism-oriented training events they had attended, the majority (56 respondents) (40.6%) responded that they attended such training events 2-4 times at maximum. It was reported that 14 respondents had attended once (10.1%), 29 respondents had attended 5-7 times (21.0%), six respondents had attended 8-10 times (4.3%), and one respondent had attended 11 times or more. Since 28 respondents had not been involved in such training before, they left the item unmarked. In addition, the remaining four respondents left the item unmarked despite the fact that they had attended such training.

4.2. Psychometric properties

It is predicted that the variables in the research scale show normal distribution separately and in combinations with all variables (Çokluk, Şekercioğlu, & Büyüköztürk, 2018:16-17). In other words, it is assumed that the reciprocal effect of the variables in the research will show normal distribution (^{Hair, Black, Babin, & Anderson, 2014}:686). Multivariate normality tests can be run graphically and statistically. One of the simpler graphical methods is to examine the histograms for each variable by plotting them on the normal curve. Q-Q plots can be shown as the more popular chart type. In this graph, the observations are ascending and graphed to correspond to the expected normal scatter plots. The values observed in this chart are on the X-axis, and the expected values are on the Y-axis. In charts with normal scatter, points are collected around a straight line. As the values (points) of the variables move away directly, the deviations from normality increase. As the points approach the line, that is, just above and below the line, it can be interpreted that the deviations from the norm are not excessive (Çokluk et al., 2018:15).

As a result of the Q-Q analysis, it can be said that our data provides normality.

In Figure 2, the distribution of normality is given by the Q-Q plots. It is observed that the data in this matrix show a close distribution to the line.

Figure 2 Q-Q plots 

Statistically, multivariate normal distribution can be assessed in the light of skewness and kurtosis values (^{Hair et al., 2014}:31-32; Çokluk et al., 2018:16; ^{Tabachnick & Fidell, 2014}:113). Levels of significance differ at critical skewness and kurtosis values which are presented below. Hair et al. (2014) indicate that the threshold value of normality should be ±2.58 at the significance level of 0.01 and ±1.96 at the significance level of 0.05. As a result of the skewness and kurtosis tests (p > 0.05), it is noted that skewness values range between -0.869 and 0.076 and kurtosis values range between -1.112 and -0.105. Since the analysis results fall into the range of the recommended values, it will undergo parametric tests in the following stages of the study.

The reliability coefficient of the scales in a study should be minimum (Cronbach’s alpha) and/or greater than 0.70 (^{Hair, Anderson, Tatham, & Black, 1998}). In other studies in the literature, 0.60 and 0.50 are also considered acceptable. In the current study, an acceptable threshold of reliability was determined to be 0.70. It is observed that the values in Table 2 (perceived overqualification: 0.862, employee performance: 0.817) are above the acceptable level of reliability.

Factor analysis is an analysis method utilized to test measurement models that bring together a large number of variables that are linked to one another, and it aims to discover a small number of conceptually meaningful constructs (factors, variables) based on the large number of variables, or to identify correlations between the factors and their indicators. Factor analysis is one of the statistical techniques used to obtain evidence for construct validity in the scope of studies in the social sciences where the main purpose is to develop or adapt a scale and in studies where an already-developed scale is used for a different purpose or on a different sample (Çokluk et al., 2018:177- 178; ^{Tabachnick &Fidell, 2014}).

Factor analysis is divided into two categories as exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) (Çokluk et al., 2018:177-178; ^{Tabachnick and Fidell, 2014}:614; Salvation, 2010:189). Although exploratory factor analysis is generally performed at the first stage of the research, it is mostly conducted to determine under which factor the measurement items accumulate. In general, it is employed in research of an exploratory nature. It is performed to explore the construct within variable sets of a newly developed measurement scale. Basically, in this analysis, it is not possible to predict in advance the number of factor groups under which the correlations between variables will accumulate. The researcher conducting the study examines the dimensions which are formed as a result of grouping the variables in line with the level of relevance in terms of content and factor loadings, renames the factors, and defines the dimensions (^{Kurtuluş, 2010}:189). At this point, the fact that the factors to be formed at this stage show consistency with the theory is of great importance for the explanatory power of the study. On the other hand, in confirmatory factor analysis, the researcher has a theoretical argument about the measures of a construct (^{Hair et al., 2014}:93). In the scope of CFA, the correlations laying out the foundation of the construct formed by the variables are put forward through validation of the theoretically developed specific hypotheses. In general, CFA is performed to identify whether the original construct of the previously used measurement scales is verified via the collected data in the study (^{Gürbüz & Şahin, 2015}:326). Whether the model formed as a result of the confirmatory factor analysis and the collected data fit one another is assessed in light of the goodness-of-fit indices. The goodness-of-fit indices of SEM and CFA in the current study are presented in Table 2.

Table 2 Goodness-of-fit indices of SEM and CFA 

The root mean square error approximation (RMSEA) is mostly the comparative fit index preferred in large samples. It is an index used to determine the population covariances in the decentralized Chi-square distribution (^{Meydan & Sesen, 2015}:34).

The comparative fit index (CFI), compared to NFI and NNFI, is a less sensitive comparative fit index. The harmony of the model is determined by comparing the model with the independence model/absence model, which assumes that there is no relationship between the variables. In other words, the independence model compares the covariance matrix and the proposed model's covariance matrix (^{Meydan and Sesen, 2015}:34).

The goodness-of-fit index (GFI) is the absolute fit index for evaluating the model fit independently of the sample size. It shows how much the GFI model, which is similar to R2 in multiple regression, measures the covariance matrix in the sample. In other words, it deals with the amount of variance and covariance that can be explained by the model (^{Meydan & Sesen, 2015}:34).

The normed fit index (NFI) is the preferred comparative fit for most large samples, and it is found by determining the ratio of the tested model to the Chi-square value of the independent model (^{Meydan & Sesen, 2015}:33).

The adjusted goodness-of-fit index (AGFI) is the absolute fit index preferred for larger samples and is a regulated form of the GFI where the sample width is taken into account. More parameters are released and based on the correction to the model's degree of freedom (^{Meydan & Sesen, 2015}:35).

The standardized root mean square residual (SRMR), in addition to being a general fit measurement based on the fit of values (such as RMR), depends on the variance and covariance sizes of the observed variables. This makes it almost impossible to decide whether the RMR value fits well or badly when the scales of the variables are not taken into account. SRMR was also developed to close this gap (^{Schermelleh-Engel et al., 2003}:12).

In the non-normed fit index (NNFI), the model's Chi-square value is added to the Chi-square ratio of the independent model and the degree of freedom (df) of the model (^{Meydan & Sesen, 2015}:33). The term TLI (e.g., in AMOS software) is also a comparative fit index used instead of the NNFI.

The incremental index of fit (IFI) uses a calculation that is essentially the same as NFI and is based on sample size and parsimony. Unlike the NFI, the degree of freedom is also taken into account (^{Byrne, 2010}:79).

In the current study, the perceived overqualification scale developed by ^{Maynard et al. (2006}) and the employee performance scale developed by ^{Kirkman and Rosen (1999}:67) and adapted by ^{Sigler and Pearson (2000}:34) were used. Since the scales in the present study were adopted from previous studies, the multi-factor first-order CFA analysis was performed to determine whether the measurement constructs are validated by the collected data. The results of the multi-factor first-order CFA are provided in Table 3.

Table 3 Results of first-order multi-factor CFA 

Following the confirmatory factor analysis, the items PO.9 (“I have more skills than required to perform my job”) and EP.1 (“I fulfill my tasks on time”) were excluded from the analysis as their factor loadings were below 0.50, which hindered the verification of convergent validity and discriminant validity. For convergent validity, the average variance extracted (AVE) and composite reliability (CR) values were tested on the basis of each latent variable. The AVE values and factor loadings ​​must be greater than 0.50 while CR values ​​must be greater than 0.70 (^{Hair et al., 2014}:619; ^{Dedeoğlu, Balıkçıoğlu, & Küçükergin, 2015}:9; Dedeoğlu, 2019:523). In SEM, the researcher predicts a fit between the theoretical argument and the data; therefore, x² is supposed to be deemed insignificant. The x² value is highly dependent on the sample, and it is generally deemed significant at sample sizes of n < 200 (^{Gürbüz and Şahin, 2015}:329). In the current study, the x²/df (1,334) and p (0.56) values ​​are at the expected levels; in other words, the model is validated by the collected data. The ranges of the other goodness-of-fit indices are provided in Table 2.

4.3. Hypothesis testing with AMOS

The main objective of structural equation modeling (SEM) is to test a set of correlations between one or more independent variables and one or more dependent variables (^{Gürbüz & Şahin, 2015}:323). The structural model developed in the scope of the current study employs one independent and one dependent variable. This model can be tested via classical methods of analysis. Nevertheless, the reason underlying the preference of SEM in the present study is that SEM, unlike the classical methods of analysis, incorporates statistical error variances into the analytical process and thus increases the explanatory power of the model. Furthermore, in classical methods of analysis that aim to test the correlations between the constructs, it is difficult to test a specific model in a holistic way. However, in SEM, whether a research model involving different correlations is holistically verified by the research data can be assessed based on goodness-of-fit indices. In the scope of the study, SEM was conducted to test the model and the hypothesis developed based on the current literature. The findings of the SEM analysis are provided in Table 4 in detail.

Table 4 Results of structural equation modeling (SEM) analysis 

As seen in Table 4, perceived overqualification affects employee performance negatively and significantly (β = -0.446; t = -3.973; p < .001). Therefore, H1 is supported in the light of this finding. In parallel with the findings in Table 4, employee performance is predicted to decrease by -45% in the event of one-unit alteration in perceived overqualification. The goodness-of-fit indices ​​provided in the structural model are a part of the goodness-of-fit indices presented in Table 2.