General digital competences of beginning trainees in commercial vocational education and training

Against the background of digital transformation processes that are currently changing the world of work, this paper examines general digital competences of beginning trainees in commercial vocational education and training (VET) programs. We are particularly interested in factors influencing digital competence profiles. From survey data including N = 480 trainees in one federal state in Germany, we were able to identify three different competence profiles (based on the trainees’ self-assessment of their general digital competence). Initial descriptive analysis reveals differences between competence profiles of different training professions (industrial clerks and retail salespersons reach higher competence levels than salespersons). However, regression results indicate that these differences can be explained by differences in school leaving certificates. Contrary to prior empirical evidence, we find no significant effect of trainees’ gender. Finally, the frequency of certain private digital activities (e.g. using office programs, conducting internet searches) affects digital competence profiles. Implications for both VET programs and further research are discussed.

digital transformation processes training companies are in need of trainees who start their training program with solid digital competences (Härtel et al. 2018). The on-going COVID pandemic additionally increases the need for digital competences, as health regulations require the use of digital tools in both the workplace (e.g. communication and collaboration) and vocational schools (e.g. distance learning formats) during the training program.
While it is often assumed that these days adolescents naturally possess certain digital competences, Kirschner and De Bruyckere (2017) illustrate that the idea of informationskilled digital natives, who readily apply technology because they grew up in a digital world, is a myth. In fact, studies among young people during general education in Germany reveal deficits with regard to digital competences (Bos et al. 2014;Eickelmann et al. 2019;Härtel et al. 2018). However, there is still a lack of empirical evidence on digital competences of adolescents at the start of VET programs (Härtel et al. 2018). Our study aims to address this research gap by examining the level of general digital competences of trainees when they enter their training program. In addition, we are interested in factors that explain different competence levels. We specifically examine the extent to which competence differences can be explained by individual characteristics as well as prior learning processes of adolescents. The study focuses on the most popular field of VET in Germany (in terms of yearly numbers of beginning trainees) and examines trainees in three commercial VET programs: industrial clerks, retail salespersons, and salespersons.

Conceptualizing digital competence
A wide range of terms are used by different authors to conceptualize individuals' abilities to use information and communication technology (ICT) (Ilomäki et al. 2016). In the following, we will refer to the concept of digital competence. Digital competence can be defined as 'confident, critical and creative use of ICT to achieve goals related to work, employability, learning, leisure, inclusion and/or participation in society' (Ferrari 2013). In a slightly broader approach, Ilomäki et al. (2016) define digital competence as consisting of '(1) technical competence, (2) the ability to use digital technologies in a meaningful way for working, studying and in everyday life, (3) the ability to evaluate digital technologies critically, and (4) motivation to participate and commit in the digital culture' . A widely used conceptualization of digital competence is provided by the European Digital Competence Framework (DigComp) (Ferrari 2013). Based on the claim that every citizen needs digital competences to participate in an increasingly digitalized society, the framework distinguishes between five areas of digital competence (Ferrari 2013): 1. Information (browsing, searching and filtering information; evaluating information; storing and retrieving information). 2. Communication (interacting through technologies; sharing information and content; engaging in online citizenship; collaborating through digital channels; netiquette; managing digital identity). 3. Content creation (developing content; integrating and re-elaborating; copyright and licenses; programming). Findeisen and Wild Empirical Res Voc Ed Train (2022) 14:2 "Computer and information literacy") based on a classic input-process-outcome model (see e.g. Fraillon et al. 2020;Heldt et al. 2020). The model assumes that input factors (antecedents) directly affect learning processes (process), learning processes, in turn, are expected to correlate with digital competence (outcome)-hence, they affect digital competences and are also influenced by competences (Heldt et al. 2020). The model also distinguishes between four levels that are relevant for the acquisition of digital competence: (1) the wider community (characteristics of the educational system, policies and curricula), (2) the school/classroom (characteristics of the school, classroom instruction), (3) the home environment (family background, e.g. migration background, and ICT access at home), and (4) the individual student (student characteristics, learning process and level of performance). The ICILS framework deliberately includes individuals' learning processes outside of school as digital competence is not only acquired in the school context. The model also accounts for the fact that antecedents and processes might be determined by factors on higher levels (e.g. ICT education policies determine schools' ICT resources) (Fraillon et al. 2020).
In our study, we are interested in the individual level, hence, in the role of characteristics and learning processes of adolescents on the acquisition of digital competence. Existing empirical evidence on the impact of individual factors is reported in the following section.

Determinants of digital competence of adolescents: empirical evidence
The International Computer and Information Literacy Study (ICILS) repeatedly reports deficits regarding digital competences among German adolescents (Bos et al. 2014;Eickelmann et al. 2019). In this comparative international study, German 8th grade students  Fraillon et al. 2020, p. 7) Findeisen andWild Empirical Res Voc Ed Train (2022)  When it comes to determining factors of digital competences, several studies have examined the effects of gender. The ICILS finds that for the German sample, girls possess significantly higher digital competences than boys. In fact, none of the other countries that are part of the study report advantages of male participants compared to females . These findings are supported by other studies that also find advantages for females regarding digital competences (e.g. Siddiq and Scherer 2019). However, there is also empirical evidence suggesting higher digital competences of males (e.g. Goldhammer et al. 2013) as well as studies finding no gender effect (Hatlevik and Christophersen 2013).
Moreover, education seems to be related to digital competence. As Hatlevik and Christophersen (2013) demonstrate for a group of students in upper secondary schools in Norway (N = 4087), the study program (vocational vs. general education) significantly predicts digital competences. The authors perceive the study program as an indicator of academic aspiration and illustrate that students in general educational tracks significantly outperform vocational track students with regard to digital competence. This replicates findings of previous studies (Calvani et al. 2012;Hatlevik 2010). For the German context, Wild and Schulze Heuling (2020) indicate that students in cooperative higher education programs demonstrate higher digital competences than students in VET. Furthermore, Hatlevik et al. (2015b) find that for a sample of ninth grade students in Norway (N = 852)-apart from family background-prior academic achievement (grades achieved in the most important school subjects) is the most important predictor of digital competence.
When it comes to the effect of learning opportunities, Zhong (2011) finds from PISA data that both ICT access at home and at school significantly predict adolescents' selfreported digital competence. However, the effect of ICT access at home is higher than the school effect and the students' previous experience in using a computer significantly predicts digital competence.
Empirical evidence on digital competences of trainees in VET is still scares. There is, however, one study from 2013 on the internet use of German trainees, finding that participants do well when it comes to navigating through the internet (orienting themselves on an unknown website, registering for a platform with their email address) (Burchert et al. 2013). With respect to those tasks, the authors found no differences between different types of VET programs (technical vs. commercial trainees). However, all the trainees experienced difficulties when it comes to searching for information and reading web content. Furthermore, the results reveal that the trainees use the internet mainly for communication and information. While there is a high affinity to use the internet for private purposes, internet use for professional reasons in the workplace is less common. This is true for the search of information and even more so for the use of internet forums, blogs, or online videos. When facing a problem, the trainees prefer consulting experienced colleagues or other trainees before trying internet searches.
Furthermore, in a small interview study among trainees in healthcare professions (N = 3), Evangelinos and Holley (2015) demonstrate-based on the DigComp framework-that trainees perceive themselves as fairly capable with respect to ICT tasks. However, their activities cover only a narrow field of technology use, mainly for private purposes (e.g. communication via social media), and they overestimate their digital competence and fail to recognize skills necessary for the workplace.

Research questions
The empirical evidence described above mainly focuses on general education programs. For the context of VET, reliable findings are missing. Moreover, in face of the repeatedly reported deficits regarding general digital competences of adolescents (in Germany), it seems worthwhile to examine digital competences of beginning trainees in VET as well as factors predicting different competence profiles. This is the purpose of our study. We focus on beginning trainees in commercial VET in Germany. The professional field of Commercial Services, Trade, Distribution, Hotel and Tourism is the largest field in terms of the number of beginning trainees in the federal state of Baden-Wuerrtemberg (2018: 11,914 beginning trainees). Within this field we focused on the three professions with the highest numbers of beginning trainees: industrial clerks [Industriekaufmann/frau] (2018: 3,219 beginning trainees), retail salespersons [Kaufmann/frau im Einzelhandel] (3,649 beginning trainees), and salespersons [Verkäufer/in] (2,575 beginning trainees). This allows us to compare trainees in three different, yet similar professions. Hence, we can, for instance, expect fairly similar interests of adolescents applying to these training programs (e.g. affinity towards the use of digital media). At the same time, these three professions typically vary with regard to trainees' characteristics, especially regarding school leaving qualification, and to some extent gender and age (see Table 7 in the Appendix). Hence, it is of interest to analyze whether trainees in these professions differ regarding digital competence and to what extend differences can be explained by individual characteristics.
This study aims to answer the following research questions: 1. Which digital competence profiles do trainees in commercial VET possess at the beginning of their VET program? 2. Which factors predict general digital competences of beginning trainees in commercial VET?
Research Question 1 focuses on the identification of heterogeneity among beginning trainees. Research Question 2 focusses on predictors of the level of general digital competence. Based on the theoretical framework depicted in Fig. 1, there are several possible predictors. This study focuses on analyzing factors on the student level. Hence, we expect trainees' competence profiles to be influenced by (1) individual characteristics and (2) trainees' learning processes related to digital activities. Regarding individual characteristics, we examine the effect of trainees' age, gender and educational qualification. Furthermore, we are interested in differences between the three training professions we include in our analysis. Among the three professions, a training program for industrial clerks is the most highly regarded. For instance, training companies typically select applicants with higher educational qualifications for this program (two thirds possess higher education entrance qualifications; see Appendix Table 7) than for the other two training programs. Our study aims to examine, whether differences between trainees in these three training programs also occur with respect to general digital competences. Also, we examine, to what extend these differences can be explained by individual characteristics.
Apart from individual characteristics, we analyze the impact of adolescents' learning opportunities related to digital activities. Since we focus on beginning trainees in the first couple of months into the training program, participants did not yet have the opportunity to significantly benefit from profession-specific learning processes in both the training company and the vocational school. Hence, we focus on general digital competences as well as learning opportunities prior to/outside of the VET program (experiences from digital activities at home or in school). As it is not uncommon to complete more than one training program, we control for previously completed training programs of trainees. In doing so, we can take into account if trainees did have access to vocational learning opportunities regarding digital activities.

Research design and instruments
We collected data from 480 trainees in commercial VET programs during their first months into a vocational training program. Data collection lasted from October 2018 to February 2019 and covered five vocational schools and 22 classes in the federal state of Baden-Wuerttemberg (convenience sampling). Participation was voluntary, and a privacy policy was adhered to. There were no incentives for participation.
During the survey, participants answered a modified instrument designed by Müller et al. (2018) with 24 items based on the DigComp framework (Ferrari 2013) to measure the following five components of digital competence: (1) Information, (2) Communication, (3) Content creation, (4) Safety, and (5) Problem solving. The items for each dimension are displayed in Table 1. For each item, the trainees indicate whether they are able to complete the task described (e.g. online transfer of money) or how they would describe their behavior (e.g. changing passwords regularly). Hence, each item is assessed dichotomously (0: I am not able to complete this task/I do not do this regularly/I do not recognize this; 1: I am able to complete this task/I do this regularly/I do recognize this). Since the questionnaire aims at a general assessment of digital competence and is designed to be applicable to a wide range of individuals, the survey participants assess their digital competence on a rather broad level. Hence, they are not asked, for instance, to distinguish between private and professional behavior.
The use of self-reports, of course, falls short of elaborate performance-based competence measures that are increasingly state-of-the-art in commercial vocational education and training research (e.g. Seeber 2016; Seifried et al. 2020). However, due to limited testing time, a thorough performance-based assessment of trainees' digital competence was not possible in this study. A self-report questionnaire had the advantages of time and cost effectiveness. Additionally, in other fields of vocational education (e.g. health care; see Holley 2014, 2015), the DigComp framework was used as well. Overall, in view of the focus of our study, this approach seems to be suitable for the assessment of general digital competences.
To evaluate the measurement quality of the used instrument, we applied the IRTbased approach by Birnbaum (1968). In detail, we used Yen's Q3-Index with a cut-off point of 0.2 to check the local independence assumption (Yen 1993). For the items Designing web applications and Programming, this assumption was violated (Yen's Q3-Index = 0.34). However, as the knowledge necessary for designing web applications is not identical to the knowledge required for programming in general, from a content point of view, we decided against excluding either of the items. All the other items were below the cut-off point (Yen's Q3-Index < 0.2). Moreover, the scales revealed fair reliability (see also  Next, we checked for the multidimensionality of the instrument by the estimation of four different models: (1) a one-dimensional 1 PL model, (2) a one-dimensional 2 PL  Table 2), and (4) a five-dimensional 2 PL model. We used the Akaike information criterion (AIC), Bayesian information criterion (BIC), Log-likelihood (LL), and Deviance to analyze model fit.
The different models analyzed and the corresponding fit indices are displayed in Table 3. Based on χ 2 -difference tests, we specifically tested the multidimensional model with five competence dimensions (2 PL model) against the three other models: (1) the model with one competence dimension (1 PL model) (χ 2 = 382.31; df = 31; p < 0.001), (2) the model with one competence dimension (2 PL model) (χ 2 = 197.87; df = 10; p < 0.001), and at last (3) the five-competence dimension 1 PL model (χ 2 = 115.72; df = 17; p < 0.001). We found that model fit was significantly higher for the five-competence dimension 2 PL model compared to all other alternative models. Hence, further analyses were based on the model with five dimensions and a 2 PL structure. Spearman intercorrelations (r s ) between the five dimensions varied between r s = 0.46 and r s = 0.69 (see Table 2). The lowest correlation existed between Content creation and Safety (r s = 0.46). The highest correlation was between Information and Communication (r s = 0.69). 1 To assess trainees' learning processes and learning opportunities (see Research Question 2), we also used an instrument by Müller et al. (2018). Here, the participants indicated which activities they performed regularly (once or several times a week). Based on face validity aspects we categorized the four items 'using searching tools on the internet to find content/information' , 'viewing online videos (e.g. YouTube)' , 'using digital maps  and route guidance systems (e.g. Google Maps)' , and 'using learning opportunities on the internet (e.g. online course, learning languages online)' as Collecting information and learning. Moreover, we clustered three items to the aspect Communication and collaboration ('using instant messaging services (e.g. WhatsApp, Threema, Telegram)' , 'using cloud services (e.g. Dropbox, Google Drive, Amazon Drive)' , and 'collaborating within a team via online tools (e.g. Google Docs, Microsoft SharePoint)'). In the category Generating content, we summarized the items 'using office programs (e.g. Word, Excel, Pow-erPoint)' and 'reading blogs and forums or creating blog entries' . The participants were asked to select all activities they regularly perform and leave unchecked the activities they do not perform (regularly) (dichotomous classification). Again, the questionnaire did not explicitly distinguish between private learning processes and work-related learning processes, however, since the trainees were only a couple of weeks/months into the training program when they filled out the questionnaire, we expect learning processes regarding digital activities to rather occur during their leisure time. Table 4 gives a summary of the sample of 480 trainees collected in the course of this study. The sample consists of 205 industrial clerks, 145 retail salespersons, and 130 salespersons. Of all participants, 61% were female, 37% male, and 2% could not be assigned to either male or female. On average, the trainees were M = 19.38 (SD = 2.35) years old. Eleven percent of the trainees successfully completed a VET program in a different profession before starting the current program. The trainees are trained in three different commercial professions: School leaving certificates varied. Almost 43% had a General Certificate of Secondary Education [Realschule]. About 20% had a lower school leaving certificate [Hauptschule]. The rest gained a higher education entrance qualification (Abitur: 18%) or technical college entrance qualification (Fachhochschulreife: 18%). 2 We found a significant difference in the training professions with respect to gender (χ 2 (4) = 13.75, p ≤ 0.01, Cramér's V = 0.12). The ratio of female trainees is higher among industrial clerks (69%) and retail salespersons (59%) than among salespersons (52%). Further analyses reveal significant differences between the trainees' school leaving certificate in different professions (χ 2 (8) = 197.67, p ≤ 0.001, Cramér's V = 0.45). An equal share (33%) of industrial clerks had a school leaving certificate at General Certificate of Secondary, advanced technical college entrance qualification, and general university entrance qualification (Abitur). Most retail salespersons had a General Certificate of Secondary Education (61%). For salespersons, there was a higher frequency of lower school leaving certifications (48%) and General Certificate of Secondary Education (40%). Finally, there were differences between the training professions in relation to former vocational apprenticeships (χ 2 (2) = 6.50, p ≤ 0.05, Cramér's V = 0.12). A successfully completed VET program was most common among salespersons (16%), compared to 12% for retail salespersons and 7% for industrial clerks. More detailed information for each professional path-also regarding digital activities and learning opportunities-is presented in Table 4.

Data analysis
In the first step, we analyzed trainees' digital competences separately for each profession. We report descriptive data based on boxplots. To test the differences of the five competence dimensions between different training professions, we used the Kruskal-Wallis-tests and the post-hoc-tests of Dunn (1964) with Bonferroni correction. Next, we applied a latent profile analysis (LPA) with the aim of grouping homogenous participants into heterogeneous groups (Oberski 2016;Vermunt and Magidson 2002). As decision criteria, we used the Aikake information criterion (AIC) and the Bayesian information criterion (BIC). In addition, we checked for the entropy values closest to 1 (Asparouhov and Muthen 2018;Celeux and Soromenho 1996), and also used the diagonal of the average latent class probabilities for most likely class membership as a selection criterion. For the latter, the cut-off criterion was an assigned class of above 80 percent (Jung and Wickrama 2008;Rost 2006).
To analyze the research questions described in Sect. 2, we used an ordinal regression (Hosmer et al. 2013). This type of regression is a sub-type of logistic regression where the dependent variable is ordered. These analyses differ with regard to calculations of probabilities. While a logistic regression provides probabilities that a variable will take on a specific value, ordered logit provides probabilities that values will fall below a certain threshold. To check the robustness of the results, we estimated nested models with 500 bootstraps. Multicollinearity was not a problem in the estimated models (VIF ≤ 1.48 for all variables used).
Five participants provided no information on socio-demographic data; these participants were excluded from the regression analysis. Apart from that, the data set contained only single missing values regarding the variables age and gender (nine missings each). Hence, we decided not to impute missing data (Tabachnick and Fidell 2013). When analyzing the effect of school leaving certificates, we excluded dropouts (n = 6) and grouped together the two different types of higher education entrance qualification (general university entrance qualification [Abitur] and advanced technical college entrance qualification [Fachhochschulreife]). We estimated the LPA using the software R with packages 'idyLPA' and 'tidyverse' . All other analyses were carried out in STATA (Version 14) and SPSS (Version 27).  Dunn (1964) was used to test differences between the three professions in detail. For the dimension Information, the results show significant differences between salespersons and retail salespersons (z = − 2.99, p < 0.01), salespersons and industrial clerks (z = 6.30, p < 0.001) Findeisen and Wild Empirical Res Voc Ed Train (2022) 14:2 as well as retail salespersons and industrial clerks (z = 3.18, p < 0.01). For the dimension Communication significant differences are revealed between salespersons and industrial clerks (z = 5.14, p < 0.001) as well as between retail salespersons and industrial clerks (z = 3.15, p < 0.01). For Content creation we again find significant differences between salespersons and industrial clerks (z = 4.99, p < 0.001) as well as between salespersons and retail salespersons (z = − 2.89, p < 0.05). Similar results are found for Safety (salespersons and industrial clerks: z = 4.82, p < 0.001; salespersons and retail salespersons: z = − 3.26, p < 0.01) as well as Problem solving (salespersons and industrial clerks: z = − 3.15, p < 0.01; salespersons and retail salespersons: z = 2.75, p < 0.05). Using a latent profile analysis (LPA), we identify profiles of digital competences. In the analysis, we test different amounts of profiles (one to five profiles) against each other. Table 5 provides the fit statistics. The results show that AIC and BIC decrease from the solution with one profile to five profiles. However, entropy suggests a solution with three profiles (highest entropy value = 0.93 for a three profile solution). The same is true for the diagonal of the average latent class probabilities for most likely class membership. The highest minimum (96%) and highest maximum (98%) are reached in the solution with three profiles. Based on these results we decided to distinguish three competence profiles for further analysis. Figure 3 depicts the three profiles for the five digital competences. The first profile (line dashed dotted) comprises 22% of the sample and shows the lowest digital competences in all five dimensions. For further analysis, we call this profile low competence level profile. A second profile (43% of the sample) achieves the highest values in all five competence dimensions. We call this profile high competence level profile. The values of a third profile (35% of the sample) lie between the two previous described profiles in all dimensions (dotted line). We name this profile medium competence level profile.

Regression results
To examine our research questions, we applied ordinal regression analyses (n = 460). Table 6 reports the regressions results. Model 1 includes trainees' professional path as well as their individual characteristics (Pseudo R 2 = 0.05; Nagelkerke R 2 = 0.10; Cox & Snell R 2 = 0.11). In this model, we find a significant effect of the trainees' profession on general digital competences. The probability of being in a higher profile of digital competences (odds ratio [OR]) is almost four times higher for industrial clerks (OR = 4.22; p < 0.001) and about twice as high for retail salespersons (OR = 2.26; p < 0.01), as compared to salespersons. Participants' gender does not significantly affect their digital competences. Furthermore, there is a marginally significant effect of the trainees' age. Each additional year increases the odds ratio to belong to a higher profile by 9% (OR = 1.09; p < 0.10). However, as the results of Model 2 indicate, the effect of both age and training profession can be explained by differences in school leaving certification. The age effect becomes insignificant in Model 2, when controlling for school leaving certificates, and the differences in probability to belong to a higher profile are reduced to OR = 2.06 (p < 0.05) for industrial clerks. A likelihood ratio test between Model 1 and Model 2 yields evidence of a modest improvement in model fit (Pseudo R 2 = 0.06; Nagelkerke R 2 = 0.14; Cox & Snell R 2 = 0.13; χ 2 (2) = 14.88, p < 0.001). In Models 1 and 2, a former degree in another VET program significantly decreases the odds to belong to a higher profile by 50 percent (Model 2: OR = 0.50; p < 0.05).
In Model 3, we included different digital activities (learning processes) (Research Question 2). A likelihood ratio test between Model 2 and Model 3, again, shows a modest improvement in model fit (Pseudo R 2 = 0.17; Nagelkerke R 2 = 0.35; Cox & Snell R 2 = 0.30; χ 2 (9) = 104.46, p < 0.001). In Model 3 (see also Table 6), the effect of the trainees' profession becomes entirely insignificant. Instead, school leaving certificates explain a significant amount of the differences in digital competence. Compared to a lower school leaving certificate, the trainees with a certificate of secondary education are twice as likely (OR = 2.01, p < 0.05) and trainees with a higher education entrance qualification are three times as likely (OR = 3.06, p < 0.01) to belong to a higher competence profile. Compared to Model 1 and 2, the effect of a former training program becomes insignificant. When it comes to digital activities and learning opportunities, for the dimension Collecting information and learning, we find a significant positive effect for 'using searching tools on the internet to find content/information' (OR = 1.98; p < 0.01) and surprisingly, a significant negative effect for 'using learning opportunities on the internet' (OR = 0.53; p < 0.05). Regarding the dimension Communication and collaboration, the item 'using instant messaging services' has a significant positive effect (OR = 2.82; p < 0.001). Finally, the items 'using office programs' (OR = 2.63; p < 0.001) and 'reading blogs and forums or creating blog entries' (OR = 2.16; p < 0.01) of the dimension generating content have a positive effect. Table 6. Ordinal regression of digital competence profile with 500 bootstraps (n = 460). 3

General discussion
The aim of this study was to examine profiles of general digital competences of beginning trainees as well as factors (individual characteristics and learning opportunities) influencing the digital competences of beginning trainees. Against the background that training companies can benefit from trainees who begin their training program with a certain level of digital competence, we claim that the competences measured using the DigComp framework form the basis for a successful start in commercial training programs and the acquisition of profession-specific digital competences during the VET program. Our analysis is based on beginning trainees in three different commercial VET programs and cannot be generalized to other VET programs. For the sample examined, we identified three different profiles of digital competence that can be characterized as low (22% of the sample), medium (35%), and high digital competences (43%). Initial results point towards significant differences regarding digital competences between different training professions. In detail, both industrial clerks and retail salespersons seem to outperform salespersons for each of the five dimensions of digital competence. However, further analysis demonstrate that these effects can be explained by differences in the trainees' school leaving qualifications. When controlling for school leaving certificates, the only effect that remains is an advantage of industrial clerks compared to salespersons. This effect also becomes insignificant when controlling for learning processes (digital activities). The finding is also in line with results from research showing that prior academic achievement is the most relevant predictor of digital competence (e.g. Hatlevik et al. 2015b).
We find no significant effect of gender on general digital competences of beginning trainees in commercial VET programs. Although most studies on gender effects pointed towards significant effects in favor of female students in general education programs, this result could not be replicated for trainees in VET. This finding might be explained by the assessment method that is based on self-reports (see also Sect. 5.2). With regard Findeisen and Wild Empirical Res Voc Ed Train (2022) 14:2 to digital competences, it is well documented in prior research that male participants report higher self-efficacy regarding advanced digital skills (e.g. Gerick et al. 2019). This bias could overshadow differences that might exist in favor of female participants. Moreover, there is no significant effect of the trainees' age, once we control for school leaving certificates. Hence, the mere age does not seem to matter with respect to trainees' general digital competence.
Furthermore, our results indicate that trainees who already participated in a prior VET program do not have a higher probability of belonging to a higher competence profile once individual learning processes are controlled for. This variable was used as a control variable to account for prior experiences regarding digital activities in vocational contexts. An insignificant effect could indicate that participants did not take their digital activities during prior training programs into account when reporting their learning processes.
Finally, our results reveal certain effects of the trainees' digital activities or (general) learning opportunities on digital competences. In line with expectations, trainees who regularly (1) use searching tools on the internet to find content/information, (2) use office programs, and (3) read blogs and forums or create blog entries reach higher profiles of digital competences. These three learning opportunities can be expected to be directly related to general digital competences. We also find a positive effect of the regular use of instant messaging services. This finding might be explained by a general affinity toward the use of digital tools and might therefore be related to general digital competences. Surprisingly, the regular use of learning opportunities on the internet is negatively related to general digital competences. This might be attributed to the fact that trainees do not perceive this activity to be relevant for the development of digital competence. However, further research is necessary to examine this relationship. Finally, the regular use of cloud services and the collaboration within a team via online tools do not significantly affect general digital competences. This finding can probably be explained by the assessment method. The DigComp framework does not account for these or similar aspects when assessing general digital competences.

Limitations and future research
This study has several limitations that need to be considered when interpreting the results. First, as already mentioned in Sect. 3.1 the use of self-reports contains certain limitations regarding the validity of digital competence assessment. A major disadvantage of self-reports is that the respondents might have distorted self-perceptions. This could lead to severe overestimations of their own abilities. However, several studies reveal that students' ICT self-efficacy positively predicts digital competence (Hatlevik et al. 2015a(Hatlevik et al. , 2018. Hence, it can be assumed that self-reports can at least be used as an indicator for actual digital competence. Moreover, our study focused on trainees' general digital competences, as we focused on beginning trainees and aimed at an assessment of their starting conditions. Also, the study focused on the sector of commercial VET. Hence, we are neither able to draw any conclusions regarding profession-specific digital competences (e.g. handling big data, using ERP systems that are especially relevant to industrial clerks; see Sect. 1.2) nor regarding other fields of VET or other training professions.