Persistence in studies in relation to learning approaches and first-year grades: a study of university chemistry students in Finland

Mika Lastusaari *a, Eero Laakkonen b and Mari Murtonen b
aDepartment of Chemistry, University of Turku, Turku, Finland. E-mail:
bDepartment of Teacher Education, University of Turku, Turku, Finland

Received 5th October 2018 , Accepted 10th May 2019

First published on 31st May 2019


Changing majors or dropping out are of great concern to universities worldwide, but the role of learning approaches in terms of students’ persistence has not been previously studied. Changing majors, especially in chemistry, is a severe problem in Finland. Here, learning approach data were collected with the ChemApproach questionnaire from 733 bachelor-level students at four Finnish universities. Students intending to change majors showed stronger submissive surface approaches and weaker active deep approaches than those intending to persist in chemistry. The ChemApproach data were complemented with information on actual persistence and first-year grades from a smaller sample from one university (N = 177). A practical deep approach in chemistry studies combined with relatively high grades was shown to be connected to persistence, while a desire to change majors combined with high grades resulted in the actual changing of majors. A high submissive surface approach indicated students at risk of dropping out completely.


Persistence in studies is of great interest to universities and their faculties, because teaching given to students who eventually drop out can be regarded as an inefficient use of resources. The percentage of students who drop out of university studies can be considerably high, and thus increasing persistence is not merely a cosmetic issue. For example, the overall dropout rate in European universities is around 30% (Vossensteyn et al., 2015; Higher Education Statistics Agency (UK), 2018); whereas in the US, an average of 60% of students drop out of bachelor's studies (National Center for Education Statistics (USA), 2017). In Finland, the overall dropout rate in universities is only 5.3%, but for chemistry education it is around 20% (Statistics Finland, 2018). The much lower dropout rate in Finland in comparison with, e.g., the US or the UK, is because all Finnish students must take an entrance exam prior to being admitted to a Finnish university. Thus, there is always an initial selection procedure that prevents the admission of students who do not have enough background knowledge of chemistry to cope with university studies. The higher overall dropout rate in chemistry, on the other hand, is a result of students using chemistry studies as a free training course for entrance exams to departments of medicine (Ruuska, 2010; Nykänen, 2013). In other words, students who have no intention of becoming chemists or physicists often enrol in chemistry or physics departments in order to gain more information on these subjects, which form an integral part of the admissions examinations for medicine. Then, after one or two years, they either drop out completely or score high enough on the admissions exams to enter medical departments. Some of these students will, however, persist with chemistry or physics. In fact, in Finnish universities that also have a faculty of medicine, the dropout rate for chemistry studies varies between 50 and 72% (Statistics Finland, 2018). This indicates that chemistry students’ attrition rate is not only of considerable importance in Finland but is also comparable with the overall dropout rates in Europe and the US.

In Finland, universities receive financial aid from the government based on their performance, one criterion of which is their output of graduated students. The funding received from the government may then be divided within a university to its faculties, and within faculties to their departments, by using the performance criteria of each unit. This means that if a student starts his or her university studies, e.g., at a department of chemistry, but graduates from medicine, it is the medical department that receives performance points, not the chemistry department. Therefore, a higher dropout rate and greater number of students switching majors will have a clear effect on the money available for a given department in a given faculty at a given university. There are also international effects, since a high university dropout rate can reflect poorly on the perceived global competitiveness of a country as a whole (Ost, 2010). The obvious goal of universities and related institutions is thus to minimise dropout rates. Accordingly, it is essential to find ways of recognising students who are at risk of dropping out as well as determining which factors affect their persistence in studies. With this information, one can begin to discuss whether universities can influence students’ persistence – and if so, with what measures.

Persistence in studies

Persistence is thought to occur in three different ways: (1) a student enrols in one programme at one university and graduates from within that same programme; (2) a student enrols in one programme, but then transfers to another programme, at the same university, graduating from the second programme; or (3) a student graduates from a university other than the one at which he or she initially began studies (Leppel, 2001). Here, we focus on the first point, i.e., with a local effect within a university chemistry department. We chose to concentrate only on this type of persistence because it demonstrates the most tangible effects for chemistry educators in higher education. In other words, it is the only persistence type that clearly influences the output of new chemistry graduates from the universities.

Until the end of the 1960s, the primary cause of student drop outs was considered to be shortcomings on the part of students themselves (Habley et al., 2012). In this regard, the first theory was proposed in 1970 by Spady (see e.g., Bean, 1982), who used the notion that shared group values and support from friends reduce suicide rates to model study persistence. Subsequently, several models have been developed to explain the factors that affect persistence or dropping out. The most influential of these models are considered to be those of Tinto (1975) and Bean (1980), both of which have been revised and augmented over subsequent years (e.g., Bean and Eaton, 2002; Tinto, 2010). Other models include e.g., those of Pascarella and Terenzini (1980): “student–faculty informal contact model”, Brower (1992): “life task model” and Altonji (1993): “sequential choice model” as well that of Cabrera et al. (1992a) that combines the models of Tinto and Bean. The earlier models have been reviewed in detail by e.g., Bean (1982), while both older and newer models were reviewed by, e.g., Kuh et al. (2006).

Tinto's (2010) model was influenced by Spady's (see e.g., Bean, 1982) and emphasises the fit between the student and his or her environment. First, the student is influenced by family, schooling and personal attributes, which together compose and direct his or her initial goals and institutional commitments. Second, during university studies, integration with the local academic and social systems refine the student's commitments further. Integration with the academic system is defined by grade performance and intellectual development, while integration with the social system comprises peer-group and faculty interactions. In Tinto's model, these refined commitments serve as the basis for choosing whether to persist or drop out. On the other hand, Bean's (Bean and Eaton, 2002) model stresses attitudes towards the intention to persist. In this model, academic, social–psychological and environmental factors, which are called socialisation and selection factors, affect grades, institutional fit and institutional commitments, thereby influencing one's decision to persist or drop out. However, in Bean's model, environmental factors (i.e., finances, opportunity to transfer and outside friends) may also directly affect one's decision to persist or drop out (Bean, 1985).

Although these models may appear conceptually straightforward, they are actually rather complex in practice. This is because they include a multitude of variables, each of which has the potential to correlate at different magnitudes and to vary in importance. Consequently, different sample groups have indicated different factors as the main contributors to persistence. These main contributors include the following: financing (Cabrera et al., 1992b; Bennett, 2003; Herzog, 2005; Lassibille and Gómez, 2008), first year grades and academic performance (Montmarquette et al., 2001; Wintre and Bowers, 2007; Araque et al., 2009; Ost, 2010; Valadas et al., 2017), major subject (Leppel, 2001; St. John et al., 2004), level of motivation (French et al., 2005) and commitment (Pascarella and Terenzini, 1983; Wintre and Bowers, 2007; Willcoxson et al., 2011), interaction (Pascarella and Terenzini, 1983; Lehmann, 2007; Wintre and Bowers, 2007), interest-major fit (Allen and Robbins, 2008), preparation for university studies (Allen and Robbins, 2008; Lassibille and Gómez, 2008), self-efficacy for educational requirements (Lent et al., 1984), major subject's first year curricula linkage (Lifton et al., 2007), student peers (Ost, 2010), parental support (Wintre and Bowers, 2007), age (Lassibille and Gómez, 2008; Araque et al., 2009), family characteristics (Lehmann, 2007; Lassibille and Gómez, 2008; Araque et al., 2009), quality of teaching (Bennett, 2003) and flexibility of studies (Di Pietro and Cutillo, 2008). As suggested from the small set of studies above, it is not enough to view persistence based only on one perspective, such as sociological, psychological, organisational, economic or cultural. Instead, an integrated viewpoint is needed (Habley et al., 2012). That said, regardless of the perspective employed, common factors are always at (inter)play between (1) who the student was before starting his or her studies and (2) how the studies influenced the student to (3) choose whether to persist or not persist. This interplay ultimately determines whether the student will persist. Here, we call these (1) pre-study factors, (2) during-study influences and (3) actualisation. A visual representation of this three-stage persistence framework in terms of the present work is provided in the Conclusions and implications.

The above-mentioned factors undoubtedly play a very important role in persistence. However, current research on student learning shows that the learning environment and the quality of learning it provokes also play a crucial role in the quality of learning outcomes and study success (Vermunt and Donche, 2017). In Finland, students comprise a homogeneous group in terms of background factors, such as economy, ethnicity and quality of prior education. This in turn presents a good opportunity to focus on the quality of learning.

Learning approaches

On the basis of a large review study, environmental factors were shown to be related to students’ learning patterns as well as to the quality of their learning (Vermunt and Donche, 2017). Additionally, students’ perceptions of their teaching–learning environment have been connected to their learning approaches (Parpala et al., 2010). Teachers’ teaching practices were shown to affect students’ learning in a study by Donche et al. (2013), in which those students whose teachers were more learning-focused in their teaching scored higher on processing strategies and better regulated their learning compared to those students whose teachers were not as learning-focused. Thus, by focusing on students’ learning patterns, we may be able to determine potential factors in the teaching–learning environment that can be changed to foster persistence in studies.

The ways in which students approach learning can be roughly described by two distant ends, namely meaning- and reproduction-directed. According to a review by Vermunt and Donche (2017), meaning-directed learning is generally positively related to academic performance, while reproduction-directed learning is typically negatively related to academic performance. Research from the Student Approaches to Learning (SAL) tradition dates back to the same time period as research on persistence. During the 1970s, pioneering works were published by four groups led by Entwistle, Biggs, Marton and Pask (see e.g., Beattie et al., 1997). The learning approach theories are based in the dichotomy of deep and surface learning: A deep approach involves the intention to understand, relating ideas and using evidence to form an overall perception; whereas a surface approach involves memorising without understanding, acceptance of not thinking and fragmented knowledge (see e.g., Entwistle and McCune, 2004). A student using the deep approach will generally consider a subject to be learned as meaningful. Moreover, such a student will feel that learning presents a positive challenge and will typically derive excitement and contentment from it. On the other hand, a student using the surface approach will most often lack these positive feelings (Howie and Bagnall, 2013). However, it has been reported that the approach to learning used by a single student is not stable but varies according to time, situation, subject matter and personal circumstances (Vermunt and Vermetten, 2004; Lindblom-Ylänne et al., 2014). One can therefore consider the possibility that common factors influence both persistence and learning approaches. Keeping in mind that research has shown that the deep approach correlates with a good learning outcome and that students can be directed towards the increased use of the deep learning approach (Baeten et al., 2010; Dolmans et al., 2010), it is compelling to think that there could be a connection between learning approaches and persistence and that persistence could be fostered by influencing the learning approach. However, the question of the relationship between learning approaches and persistence has not yet been studied extensively. In fact, to the best of the authors’ knowledge, only one publication exists on this topic. In that work, it was reported that for law students, the features of the surface approach or those of an unorganised deep approach were the most important predictors of whether a master's degree was completed within seven years (Haarala-Muhonen et al., 2017). However, no indication was given as to whether the results were due to delays in studies or to dropping out. The lack of reports on the relationship between learning approaches and persistence may be due to the fact that both may be influenced by a great number of different factors (as discussed above), and thus there may be too many variables to be analysed. Fortunately, Finland has a rather distinct educational setup with a very homogeneous group of students (when concerning, e.g., economy, ethnicity and quality of prior education). Such a high degree of homogeneity reduces the number of potentially influencing factors, making the combined study of persistence and learning approaches more feasible.

As learning approaches have been found to be context-specific (e.g., Vermunt and Donche, 2017), we have developed a questionnaire directed especially at chemistry learning: the ChemApproach questionnaire (Lastusaari et al., 2016). In our earlier study (Lastusaari and Murtonen, 2013), we found that with chemistry students, we could separate different types of deep and surface approaches. What was especially interesting was that those students aiming to persist in chemistry rated high on the active deep approach scale, which emphasised the role practical laboratory exercises play in chemistry learning. This finding is congruent with the results for pharmacy students by Smith et al. (2007), who found that those students who used a strong application-directed learning approach also demonstrated significantly positive academic performance. Another important finding from our previous study that should be investigated further (Lastusaari and Murtonen, 2013) was that surface orientation was correlated with the willingness to change majors. To obtain a more coherent picture of chemistry students’ learning and persistence in studies, we demonstrated that the questionnaire yielded valid results (Lastusaari et al., 2016). In the present work, we inspect more closely the interplay between learning approaches, persistence and grades.

Research questions of the present study

The aim of the current study was to explore how persistence in the chemistry major and learning approaches in chemistry are connected. We also analysed how course grades were connected to learning approaches and persistence. In addition, we sought to generate more information about those students who drop out from their studies. The following research questions were posed:

(1) Do students who want to persist in the chemistry major express different kinds of learning approaches than those students who want to change their major?

(2) What kinds of learning approaches are expressed by those students who are successful in changing their major?

(3) Are course grades connected to learning approaches and, further, to changing majors or dropping out?

(4) Can we predict who will change their major, and is it possible to increase the number of students who persist in the chemistry major?

Material and methods


The ChemApproach questionnaire (Lastusaari et al., 2016) was chosen for this study because it was developed specifically for chemistry and the data obtained from it were found to be both reliable as well as to give valid results. The questionnaire is based on the earlier SAL (Students’ Approaches to Learning) tradition questionnaires, e.g. Inventory of Learning Styles in Higher Education, ILS, (Vermunt, 1994), Study Process Questionnaire (SPQ) by Biggs, the Approaches to Study Skills Inventory for Students (ASSIST) by Entwistle (see e.g.Lovatt et al., 2007), Learning and Study Strategies Inventory (LASSI) and the Motivated Strategies for Learning Questionnaire (MSLQ) (see e.g., Zeegers, 2001). ChemApproach measures four features of learning approaches tailored for chemistry learning: submissive surface (SubSurf), technical surface (TecSurf), active deep (ActDeep), and practical deep (PraDeep) (Lastusaari et al., 2016). Appendix 1 provides more information on the items contained in each of the approach scales. These features can be described as follows: a pure submissive surface approach means studying chemistry in a passive way by simply memorising information. The learner is neither interested in learning nor willing to put in the effort required for learning; alternatively, the learner may not believe in his or her ability to learn chemistry. A pure technical surface approach reflects more active learning, whereby the student more strongly believes in his or her ability to learn chemistry. In this approach, the learner memorises information by actively using different memorisation techniques. In a pure active deep approach, a student seeks to actively add information to complement his or her learning. The student processes this information cognitively to form an overall picture of what must be learned. Finally, in a pure practical deep approach, a student emphasises especially chemistry laboratory work as the means to learn and truly understand the theoretical basis of chemistry. Although each of these four approaches are distinct, no student will employ any of them in isolation but will rather combine them to some degree. The intent to persist or change majors was measured in the ChemApproach questionnaire with the self-report questions, ‘Would you like to change your major subject?’ and ‘If so, to which major would you like to change?’.

Data collection

Altogether, there are six universities in Finland that offer chemistry bachelor's and master's education. Our data were collected from four of these universities, excluding the largest and smallest universities. Data were collected from students pursuing chemistry as either a major or a minor. The learning approach data were collected by teachers during lectures and laboratory work with the Finnish version of the ChemApproach questionnaire, which has been shown to yield valid results with chemistry students (Lastusaari et al., 2016). Approximately 98–100% of all students present in the data gathering situations in all universities and in each year answered the questionnaire. All data collection, handling and analysis were completed according to the Finnish ethical principles of research issued by the National Advisory Board on Research Ethics (2009). All participating students were volunteers, and all gave their written consent to participate. Each database for empirical study was processed in such a way that the participants could not be identified at any stage of data handling or from the published results. At the University of Turku, data were collected over four years, but only once from each individual student. In other participating universities, data collection was carried out for one year only. The fact that no retests were conducted may raise some concerns about how temporal stability could have affected the reliability of the results (Arjoon et al., 2013). It must be noted that the questionnaire was always administered at a pivotal point for first-year students, i.e., in late autumn, just before Christmas break. By that time, the students would have become accustomed to studying at the university and would have thus reached a maturation point in that respect. Therefore, it was not justifiable to collect data earlier. On the other hand, beginning in January, most first-year students who intend to change their major will stop attending lectures and laboratory exercises and start focusing instead on the entrance exam for their new major. Thus, it was not sensible to administer a later retest either. However, although the temporal stability of the students’ replies could not be directly investigated by retesting, the results presented later in this work nonetheless strongly indicate a high degree of temporal stability, as evident from the connections between the learning approach features and actual persistence.


The total number of participating students was 733, with the size of each dataset from each participating university as follows: University of Turku, 509; University of Jyväskylä, 116; University of Oulu, 67; and University of Eastern Finland, 41. Of these, 50% were chemistry majors, and 53% were female. Further, 25% of the students wanted to change their major, while 68% did not (7% gave no answer). In this paper, we concentrate on the chemistry majors; chemistry minors are used only for comparison. Of the chemistry majors (N = 343; note that this number is less than 50% of 733 because not all students gave information on their desire to change majors), 34% wanted to change majors, and 80% of these students wanted to change to medicine. The next most popular subject to change majors to was biochemistry (8%). This dataset of chemistry majors was later called ‘all majors’. The all majors dataset comprised 237 first-year students, and thus our sample represents 60% of the first-year chemistry major population at the participating universities and 38% of the total Finnish first-year chemistry major student population (see Appendix 2 for more detailed information).

A second dataset was subtracted from the whole dataset to include only bachelor-level chemistry majors from one university. This group (N = 177) comprised 51% females, and 41% of all students in this group wanted to change their major. This dataset was complemented with information on actual persistence (i.e., whether the students had persisted in chemistry, changed their major or dropped out) and first-year grades received through the student register.


Data analyses included descriptive statistics, confirmatory factor analyses (CFA) and analyses of variance (ANOVA), with post hoc and non-parametric analyses when appropriate. After careful data screening and descriptive statistical analyses, the structural validity of the measurement scale was evaluated. The hypothesised four-factor model was based on the results of our previous work (Lastusaari et al., 2016) and was fitted to the data using confirmatory factor analysis (CFA, see e.g.Brown, 2006). In CFA, a pre-specified factor model is constructed, then fitted to the data and evaluated. The CFA thus functions as a refinement of the factor model. The specification of the model is strongly based on theory or prior research evidence. The number of factors, the pattern of factor loadings and error theory are all specified in advance. Here, the hypothesised model was the previously validated four-factor model (Lastusaari et al., 2016) that was fitted to the data using CFA. The model was estimated using a full information maximum likelihood method with robust standard errors (MLR), which can handle non-normality, missing at random data and is recommended for small- to medium-size samples (Muthén, 2002). The model fit was evaluated using a chi squared (χ2) goodness of fit test and different kinds of model fit indices, including incremental fit indices, the comparative fit index (CFI) and the Tucker Lewis Index (TLI), which has a higher penalty for model complexity than does the CFI, as well as absolute fit indices, standardised root mean square residuals (SRMR), which have no penalty for model complexity, and root mean square error of approximation (RMSEA), which penalises for model complexity. Such a high number of different index values were considered in order to avoid making an inaccurate conclusion about model fit (Wang and Wang, 2012). A nonsignificant result of the χ2 test indicated a good fit. There are well-known problems concerning the χ2 test, such as sensitivity to sample size, an excessively stringent null hypothesis, possible problems with χ2 distribution, and the effect of the number of observed variables (Marsh et al., 1988; MacCallum et al., 1996; Byrne, 2012). Thus, the model fit evaluations were mostly based on the other fit indices listed above. The cutoff values used for accepting a model were CFI and TLI above 0.90, and RMSEA and SRMR below 0.08 (Hu and Bentler, 1999; Little, 2010; Wang and Wang, 2012). The possible need for model modifications was investigated using information from the model modification indices, which could be computed for each fixed and constrained parameter in the model. If the goodness of fit of the initial model is not satisfactory, then the model modification indices can be used as diagnostic statistics to determine to what degree the fit of the estimated model could be improved by freeing some fixed parameter.

All comparisons between the subgroups were analysed using the mean score variables, which were computed based on the factors of the CFA model. The internal consistency of these scales was evaluated using the Cronbach's alpha reliability coefficient. In the group comparisons, the between-within ANOVA was used as an omnibus test for the interaction and main effects. In case of a significant interaction effect, simple main effect analyses were conducted using one-way ANOVA with pairwise comparisons via Tukey's post hoc tests. In case of non-normality or small group sizes, the Kruskal–Wallis non-parametric ANOVA with related Bonferroni-adjusted pairwise comparisons were applied. Besides the significance tests, the effect size was evaluated with the eta squared (η2) or partial eta squared (ηp2) measurements, with the following cutoff values: 0.01 weak, 0.06 medium and 0.14 strong (Cohen, 1988).

Descriptive statistics, analyses of variance and non-parametric analyses were carried out using IBM SPSS Statistics 24.0 software. CFA was performed using the Mplus 7.4 statistical programme (Muthén and Muthén, 2015).

Results and discussion

Desire to change major as a factor influencing learning approaches

First, the complete dataset (N = 733) was subjected to CFA to confirm the functionality of the factor structure, which has been validated before (Lastusaari et al., 2016) and is discussed above. The distributional properties, number of missing data and possible outliers or extreme cases were examined using descriptive statistics and by a visual inspection with histograms, normal probability plots and box-plot graphs. Accordingly (Curran et al., 1996; Kline, 2016), the data seemed appropriate for further analyses. The results of the descriptive statistics for the observed variables and the associations between them are tabulated in Appendix 3. The reliability and the model fit were at an acceptable level as measured by Cronbach's alpha (SubSurf: α = 0.80; TecSurf: 0.64; ActDeep: 0.70; PraDeep: 0.67) and the model fit indices (CFI, TLI, RMSEA and SRMR; see Fig. 1 and Table 1). It must be noted that some of the α values fell somewhat short of the conventional 0.70 limit. However, lower values, e.g., between 0.60 and 0.70, can be considered acceptable when factor analysis results support the soundness of the model (Taber, 2017). Also, the correlation within three pairs of items must be accounted for in order to achieve this fit. However, these paired items describe rather similar attributes and can thus serve as a test of the internal consistency of a student's replies. Therefore, their pairing is supported by the learning approach theory. Thereafter, the same analysis was carried out for ‘all majors’ (N = 343) – and again, the fit indices showed that a reasonable solution was obtained (Table 1). In both the models above, the χ2 test showed significant deviation from the exact fit. However, because of the limitations of the χ2 test, its significance alone should not be a reason to reject a model (Wang and Wang, 2012); rather, the evaluation of the model should be based on other index values, such as CFI, TLI, RMSEA and SRMR (Byrne, 2012).
image file: c8rp00244d-f1.tif
Fig. 1 Standardised coefficients for the CFA model (N = 733). Latent constructs are shown in ovals, and observed variables are shown in rectangles. All coefficients except ‘ns’ are significant at p < 0.05.
Table 1 Model fit information of all the datasets. All models were four-factor CFA
Note: the cutoff values used for accepting a model were CFI and TLI above 0.90, RMSEA and SRMR below 0.08 (Hu and Bentler, 1999; Little, 2010; Wang and Wang, 2012).a The degrees of freedom are different compared to the model with the complete sample (N = 733), because one significant error covariance was added (between the items ActDeep1 ↔ ActDeep4).
Complete sample (N = 733)
110 282.60 <0.001 0.93 0.91 0.05 0.06
Subsample1: chemistry majors (N = 343)
109a 201.29 <0.001 0.92 0.90 0.05 0.06
Subsample2: one university (N = 177)
109a 143.98 0.014 0.94 0.92 0.04 0.075

Using the confirmed factor structure (see model fit information in Table 1), we computed the corresponding mean score variables and examined whether the desire to change majors would be reflected in these mean scores. The two-way interaction effect between the learning approaches and the subject of the changed major was significant (F(3, 978) = 5.89, p = 0.001, ηp2 = 0.018), and the effect size was small. Thus, simple main effects analyses were conducted for the differences between the ‘change major subject’ groups. The results (Table 2) show that the submissive surface (SubSurf) score was statistically significantly higher for students who intended to change majors. Furthermore, active deep (ActDeep) scores were statistically significantly higher for those students who did not want to change their major than for those who did want to change their major. The η2 values indicate that the effect size was medium (0.06 ≤ η2 < 0.14) for SubSurf and small (0.01 ≤ η2 < 0.06) for ActDeep. We also checked the data for students from majors other than chemistry to determine how the intention to change majors was connected, overall, to learning approaches, but we found no main effect of group (i.e., whether a student intends or does not intend to change majors) across all learning approach dimensions or factors (Appendix 4). This indicates that students who had chemistry as their minor employed the same learning approach features regardless of whether they wanted to change their major. This may be because the ChemApproach data only address the students’ learning approaches towards chemistry, not towards the subject of their major. One can assume that most of the students who had chemistry as a minor chose to take chemistry courses voluntarily. In other words, these students wanted to learn more about chemistry regardless of their commitment to their own majors. Therefore, it seems that the higher submissive surface and lower active deep scores were characteristic of chemistry majors who wanted to change their major. If we think about these results from the point of view of persistence theory (Tinto, 2006; Tinto, 2010), it can be stated that the students with high submissive surface and low active deep scores had low initial commitment toward chemistry. On the other hand, those students who indicated not wanting to change their major scored high in active deep (ActDeep). Of course, wanting to change majors is not the same as actually changing majors. There is always the possibility that one's plans will change or that these plans will not ultimately be realised. According to Tinto's (2010) model, one's goals and commitments will be refined during university study as a result of the degree of success achieved in integrating with the academic and social systems. It is only after this refinement has occurred that the decision about whether to persist is made. For this reason, we next discuss the factors that affect actual persistence.

Table 2 Means, standard deviations and one-way analyses of variance for the effects of intending and not intending to change majors on the learning approaches of chemistry majors (N = 343)
Intends to change major (N = 117) Does not intend to change major (N = 226) F(1, 341) p η 2
SubSurf 2.61 0.87 2.41 0.81 4.84 0.029 0.014
TecSurf 3.31 0.86 3.18 0.85 1.81 0.179 0.005
ActDeep 2.71 0.87 2.97 0.86 7.57 0.006 0.021
PraDeep 3.27 0.85 3.42 0.83 2.48 0.116 0.007

Learning approaches and persistence

To study whether learning approach features could be connected to students’ persistence in their initially chosen major (i.e., chemistry within one university), we used one subgroup from the whole dataset. This consisted of ChemApproach data from one university (N = 177), which were complemented with information about persistence across four years and first-year chemistry grades. First, we tested whether these data from one university were statistically significantly different from the rest of the data – no differences were found in either the means of the sumscore variables or in the desire to change majors. Next, this dataset was subjected to CFA, which confirmed that the structure of the factor model was also valid in this data subset (Table 1). The data were then grouped according to intentions to change majors (as indicated by the student in the ChemApproach form) and actual persistence information obtained from study records. The groups were as follows: (1) intended and changed (24% of students), (2) intended but did not change (23%), (3) did not intend but changed (10%), (4) did not intend and did not change (38%) and (5) dropped out (5%). Here, dropping out means that the student discontinued studies in the original university. Of those who did change their major, 82% changed to medicine.

In the between-within ANOVA, the learning approach × persistence group interaction effect was significant (F-test with Greenhouse and Geisser correction: F(11.40, 458.99) = 3.52, p < 0.001, ηp2 = 0.081), and the effect size was medium. Thus, simple main effects analyses were conducted for the differences between the persistence groups. The results show that submissive surface (SubSurf) and practical deep (PraDeep) have statistically significant differences between the five groups of persistence (Table 3, see Appendix 5 for the post hoc tests). The corresponding η2 values indicate that the effect size was medium (0.06 < η2 < 0.14) for both SubSurf and PraDeep. Because of the small number of students especially in the group of drop outs, we also carried out a nonparametric Kruskal–Wallis H test (Appendix 6), which agreed with ANOVA. In the case of submissive surface, the statistical significance was obtained for the pairs (i) ‘dropped out’ – ‘intended and changed’ (p = 0.002), (ii) ‘dropped out’ – ‘did not intend but changed’ (p = 0.004) and (iii) ‘dropped out’ – ‘did not intend and did not change’ (p = 0.005). In each case, those who dropped out scored highest in submissive surface, suggesting that this approach could be used as an indicator of students at risk of dropping out. It is also worth noting that those who intended but did not change majors did not score statistically differently in submissive surface than those who dropped out. This may be because both of these approaches can be considered as failures in studies and may thus be due to a lack of motivation or commitment or to a bad interest-major fit. All of these are important factors in persistence (Pascarella and Terenzini, 1983; French et al., 2005; Wintre and Bowers, 2007; Allen and Robbins, 2008). Students with insufficient motivation or commitment or unsuitable interest-major fit would thus naturally not persist. However, in the present case, not having enough motivation or commitment to the desired major (instead of chemistry) actually increased persistence with chemistry, even though the interest-major fit would not be optimal. Tying in with the general concepts of persistence models (Kuh et al., 2006; Tinto, 2010; Habley et al., 2012) it is possible that those students who dropped out experienced a negative during-studies influence on their pre-study factors. On the other hand, those who intended to change majors yet persisted were influenced by chemistry studies positively enough to persist regardless of their initial goal and commitment to majoring elsewhere.

Table 3 Means, standard deviations and one-way analyses of variance for the effects of intentions to change and actual persistence on the learning approaches of chemistry majors. Note that here, N < 177, because not all students gave replies to all items
Intended and changed (N = 40) Intended but did not change (N = 38) Did not intend but changed (N = 16) Did not intend and did not change (N = 63) Dropped out (N = 8) F(4, 160) p η 2 Non-parametric Kruskal–Wallis
SubSurf 2.17 0.63 2.53 0.72 2.11 0.66 2.28 0.72 3.19 0.59 4.96 0.001 0.110 19.17 0.001
TecSurf 3.29 0.74 3.25 0.93 3.49 0.85 3.14 0.78 2.98 1.10 0.83 0.511 0.020 2.83 0.586
ActDeep 2.77 0.71 2.78 0.80 3.25 0.95 2.98 0.79 2.41 0.63 2.25 0.066 0.054 8.85 0.003
PraDeep 2.93 0.64 3.12 0.75 3.19 0.65 3.48 0.74 3.43 0.35 4.34 0.002 0.098 15.87 0.003

For practical deep (PraDeep), a statistically significant difference was obtained for the pair ‘intended and changed’ – ‘did not intend and did not change’ (p = 0.001). The result is such that the latter group scored higher than the former. Therefore, a low practical deep score seems to reveal those students who will succeed in changing their major. The group of dropouts also showed an interestingly high practical deep score. It can be speculated that because chemistry is very much an empirical science, the practical deep score can be considered a good indicator of a student's interest in chemistry. One must keep in mind, though, that a low practical deep score would not mean that the student could not have a practical deep approach towards some other discipline. Even though practical deep is not an indicator of commitment as such, it could thus serve as a feature for enhancing the refinement of commitment and goal setting towards persisting with chemistry. Therefore, combining this practical deep result with the submissive surface score of the dropouts, one is tempted to wonder whether the dropout rate could be lowered by influencing the submissive surface features of these students towards a more positive, less submissive approach, one which would enhance commitment and thus lead to persistence.

First-year grades and persistence

Grade performance is one of the key features of integration with the academic system, and thus it can be considered a major factor in enhancing persistence. Indeed, first-year grades have been reported in many works to be one of the key factors for increasing persistence in studies (e.g., Montmarquette et al., 2001; Wintre and Bowers, 2007; Araque et al., 2009; Ost, 2010). Thus, we checked whether our data would show similar results. In the present study, grades were based on test scores only. Moreover, grades were given only for lecture courses, not for laboratory courses.

The results of the one-way ANOVA showed a significant main effect of the persistence group (F(4, 37.27) = 3.23; p = 0.023; η2 = 0.081), and the effect size was medium. The robust Brown–Forsythe ANOVA was used because of the unequal variances between the groups. The pairwise comparisons (Games–Howell post hoc tests) indicated a statistically significant difference in grades between ‘intended and changed’ and ‘intended but did not change’ (p = 0.001; see Appendix 7 for the complete post hoc comparison data). That is, of those students wishing to change their major, those who did change scored clearly higher than those who did not. It is important to note here that it was not the drop outs who had the lowest grades. Therefore, it was not the grades alone that affected persistence in the initially chosen major. This is a good example of academic integration being more than good academic performance. This is very well described with the concept of interest – major fit (Allen and Robbins, 2008). Again, because of the small number of students in the dropout group, we also carried out nonparametric testing. The non-parametric Kruskal–Wallis test gave a similar result (H(4) = 13.63, p = 0.009; please see Appendix 8 for the post hoc data).

Combined effects of approach and grades

Based on the results presented above (summarised above in Fig. 2), the most important feature for predicting students at risk of dropping out is submissive surface, i.e., a high submissive surface score indicates a high risk. If one considers only those who continued at university by either persisting with chemistry or changing their major, then it seems that grades and the practical deep approach are important (Fig. 3).
image file: c8rp00244d-f2.tif
Fig. 2 Mean values of the learning approach scale scores and first-year grades for students in the different persistence groups.

image file: c8rp00244d-f3.tif
Fig. 3 Mean values of the learning approach scale scores and first-year grades for students in the different persistence groups.

To examine this more, we conducted a binary logistic regression analysis (Table 4) to determine how learning approaches and grades predict persistence. According to the results, the most powerful factor for explaining persistence in the chemistry major is the practical deep approach. In this analysis, we ignored the students’ self-reported intent to change majors. Thus, on the level of examining all those who persisted or changed majors, grades did not play a central role. However, when combined with students’ own will to change majors, as seen earlier in Fig. 2, grades did play a crucial role, separating those who intended to change majors and were successful in doing so from those who intended to change majors but were not successful in doing so.

Table 4 Logistic regression predicting persistence
Predictor B SE Wald statistic p OR
Note: B = regression coefficient, SE = standard error, OR = odds ratio.
SubSurf 0.60 0.29 4.21 0.040 1.82
TecSurf −0.42 0.24 3.07 0.080 0.66
ActDeep −0.14 0.26 0.28 0.599 0.87
PraDeep 0.99 0.29 11.48 0.001 2.68
Grade −0.42 0.20 4.46 0.035 0.66

Conclusions and pedagogical implications

Our goal in this study was to explore the relationship between Finnish university chemistry students’ persistence in the chemistry major, their approaches to learning chemistry, and their grades. For persistence, we studied both their self-reported desire to persist with chemistry or their desire to change to another major, as well as their actual persistence measured one year after filling out the ChemApproach questionnaire. The ChemApproach questionnaire (Lastusaari et al., 2016) yielded reliable data for measuring students’ approaches and demonstrated clear connections between the studied variables.

According to the results, the practical deep approach to studying was the most powerful indicator for persistence in the chemistry major. Rating high on the submissive surface approach was connected to the willingness to change majors, and it also predicted dropping out totally from university. Grades did not play an important role in persistence when examining all students who persisted or changed majors; but when examining the more detailed groupings, grades did reveal that those who intended to change majors and succeeded in doing so had higher grades than those who intended to change majors but were not successful in doing so. It seems that a pre-study intention to not persist or to persist with chemistry has only a moderate effect on actual persistence, since only one-half of those who intended to change majors were successful in doing so. Also, there was a small proportion of those who did not plan to change majors but ultimately did so. If one outlines the results in the general three-stage framework common to the models of persistence discussed in the Introduction (summarised in Fig. 4), the during-studies influence, i.e., the learning process, is clearly most important. However, because the data were collected by teachers, it may be that some students felt pressure to give a socially desirable reply to the question of intention to change majors. This could be one explanation for why the pre-study factors had no effect on persistence.

image file: c8rp00244d-f4.tif
Fig. 4 The findings of the present work inserted into the general three-stage framework common to models of persistence.

The results reveal the great importance of integration to the academic system, i.e., how well the students adhere to the structural rules and requirements of the institution (Habley et al., 2012). This is reflected well in the submissive surface score. To decrease submissiveness (SubSurf score) and, at the same time, promote deep learning, one could engage in more interactive teaching methods, e.g., collaborative problem-solving practices and real working life cases, to better immerse these students in the studied themes. Donche's et al.'s (2013) study showed that teachers who were more learning-focused helped their students to score higher on processing strategies and to better regulate their learning. Similarly, changing teaching methods and the learning-teaching environment to support the development of deep learning approaches could both reduce the willingness to change majors and the likelihood of dropping out.

The persistence of chemistry majors’ interest in chemistry laboratory exercises in terms of the practical deep approach should definitely be fostered on the basis of these results. Application-directed learning has been shown to have a significant and positive relationship with academic performance in a study with pharmacy students (Smith et al., 2007). Thus, the number of practical exercises in education should not be reduced, as many contemporary educational reforms tend to do – on the contrary, they should be increased. Also, other ways to make study content more ‘concrete’ should be considered. Doing so might motivate some students to change their plans to change majors and instead persist in chemistry. This would likely also help surface-oriented students, especially those in danger of dropping out, to develop their learning approaches to be deeper-oriented. One way to develop teaching in this regard could be to incorporate smartphones or tablets. For example, Wijtmans et al. (2014) reported that connecting theory to practice, engaging students with chemistry research and maintaining student attention was improved by using multiple-choice questions, open-ended questions and molecular visualisation with mobile electronic devices as interactive methods during lectures.

It seems that the end result of getting rather high grades is driven by the fact that chemistry is one of the subjects that needs to be mastered in the admission examinations of the departments of medicine. Thus, the students with high grades are, on average, such who strongly strive towards the medical department. It would seem difficult to influence the intentions of these students, because in Finland, medical doctors not only have a higher societal status than chemists but also have considerably higher salaries. However, it may be possible to influence the practical deep score by linking first-year lecture teaching more tangibly with laboratory exercises. In addition, building better linkages between the subject matter taught and working life options could increase interest towards chemistry and thus increase persistence.

The present study shows that the ChemApproach questionnaire can be used to identify chemistry students at risk of not persisting. Thus, the questionnaire may serve as a tool for assessing the need for changes in teaching methods in order to minimise the risk of not persisting with chemistry as a major. Environmental factors have been proven to be influential in shaping students’ learning patterns (Vermunt and Donche, 2017); thus, the teaching–learning environment should be further developed in chemistry to support the practical deep learning approach and to reduce the surface approach. This study also showed that the ChemApproach questionnaire yields reliable data about students’ learning approaches, which were shown to be tightly connected to their persistence in studies. However, it must be noted that the present work was carried out using the Finnish version of the questionnaire and that no testing with the English translations presented in Appendix 1 has been done, yet.

Limitations of the study

Dropping out is a serious problem in universities worldwide. The results presented here shed new light on both who drops out and how dropping out is connected with learning approaches. We claim that the results presented here provide new and helpful information that can be put into wider use. This is especially so, because with our special case of chemistry battling against medicine, we feel that new information was obtained considering persistence in higher education. However, it must be pointed out that since Finland has a distinct educational setup with a homogeneous group of students in terms of background factors, such as economy, ethnicity and quality of prior education, it may be that the results presented here are not directly applicable to other countries, setups or disciplines. However, if the setting is similar, i.e., students must apply to university education and may not enter the programme they originally intended on, then these results are likely applicable regardless of discipline. The connection between low motivation towards a major and surface-level learning approaches likely exists in many cases in which students are not motivated to pursue their current major. On the other hand, the results regarding persisting for chemistry students who are motivated by laboratory work may be applicable to other disciplines, including via the hands-on element. We also believe that because the ChemApproach questionnaire is a simple tool, it can be easily used in different contexts to collect data and analyse institution-specific factors contributing to dropping out.

Finally, we must emphasise that although the learning approaches discussed here are a product of a complex mixture of factors that may be difficult to change, evidence does exist that they can be influenced (see e.g., Parpala et al., 2010; Donche et al., 2013). With the ChemApproach questionnaire providing clues about which learning approach features require changes, we are one step closer to improving persistence.

Conflicts of interest

There are no conflicts to declare.

Appendix 1: statements associated with the items of the ChemApproach questionnaire

Tables 5, 6, 7, 8, 9, 10, 11 and 12
Table 5 English translations of the statements associated with the items of the ChemApproach questionnaire
Item Statement
SubSurf1 Many things that I learn remain isolated and do not link as a part of a bigger picture.
SubSurf2 When reading the course material, I often do not understand how a new topic relates with any old one.
SubSurf3 I have to memorize things without having the opportunity to understand them.
SubSurf4 During a chemistry lecture, I often do not understand what a new thing is connected with.
TecSurf1 I underline while reading for chemistry examination.
TecSurf2 I divide the course material to parts, which I learn for the chemistry examination.
TecSurf3 When reading for a chemistry examination, I try to make summaries of different unities with my own words
TecSurf4 I make my own notes when studying for an examination.
TecSurf5 I make mnemonics to learn things better.
ActDeep1 After a chemistry lecture, I often chew over the things taught.
ActDeep2 I usually search and read additional material concerning the course.
ActDeep3 I often chew over the thoughts awoken by scientific texts as well as connections between them.
ActDeep4 I look for justifications and evidence to make my own conclusions about things to be learned.
PraDeep1 I like to do practicals.
PraDeep2 I have often understood a chemical phenomenon only after doing practical work on it.
PraDeep3 When doing a practical, I usually try to understand what its different parts are based on.
PraDeep4 One can learn a chemical phenomenon only by doing practical work on it.

Appendix 2: information on the annual intake of chemistry majors in all Finnish universities giving chemistry bachelor and master education

Table 6 Information on the annual intake of chemistry majors in all Finnish universities giving chemistry bachelor and master education. Information is also given on the amount of first year students participating in the present study
University Annual intake of chemistry majors, n Years collected Participating students per year, n Participation, %
Turku 63 3 59 94
Jyväskylä 55 1 17 31
Eastern Finland 28 1 14 50
Oulu 53 1 29 55
Helsinki 106 0 0 0
Åbo Akademi 6 0 0 0
All participating 199 119 60
All together 311 119 38

Appendix 3: zero-order correlations and descriptive statistics of the items and the mean score variables

Table 7 Zero-order correlations and descriptive statistics of the items and the mean score variables (N = 733)
Items 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
1. SubSurf1
2. SubSurf2 0.51
3. SubSurf3 0.50 0.52
4. SubSurf4 0.42 0.59 0.44
5. TecSurf1 0.03 0.11 0.05 0.06
6. TecSurf2 −0.10 −0.01 −0.04 −0.03 0.14
7. TecSurf3 −0.10 0.01 −0.09 0.02 0.20 0.34
8. TecSurf4 −0.01 0.08 0.03 0.05 0.41 0.26 0.38
9. TecSurf5 0.04 0.01 0.01 −0.04 0.21 0.19 0.20 0.33
10. ActDeep1 −0.14 −0.14 −0.10 −0.12 0.02 0.15 0.22 0.15 0.09
11. ActDeep2 −0.10 −0.05 −0.12 0.00 0.08 0.10 0.14 0.13 0.06 0.30
12. ActDeep3 −0.21 −0.13 −0.14 −0.11 −0.01 0.11 0.20 0.08 −0.05 0.42 0.37
13. ActDeep4 −0.15 −0.11 −0.13 −0.14 0.07 0.12 0.21 0.07 0.09 0.29 0.35 0.49
14. PraDeep1 −0.22 −0.16 −0.20 −0.18 0.04 0.10 0.09 0.04 0.08 0.17 0.09 0.17 0.12
15. PraDeep2 0.04 0.11 0.08 0.08 0.13 0.16 0.12 0.18 0.18 0.11 0.11 0.06 0.06 0.34
16. PraDeep3 −0.24 −0.21 −0.24 −0.24 0.01 0.23 0.24 0.13 0.14 0.31 0.27 0.33 0.30 0.42 0.32
17. PraDeep4 0.05 −0.07 0.07 0.10 0.10 0.12 0.09 0.07 0.07 0.08 0.11 0.09 0.08 0.23 0.51 0.22
M 2.58 2.30 2.51 2.32 2.52 3.37 3.11 3.50 3.42 3.03 2.52 2.89 2.94 3.74 3.24 3.37 2.72
SD 0.96 0.93 1.03 0.92 1.49 1.15 1.25 1.32 1.22 1.00 1.11 1.16 1.09 1.08 1.08 1.03 1.14
Skewness 0.30 0.59 0.42 0.56 0.43 −0.40 −0.12 −0.49 −0.43 −0.21 0.35 0.06 −0.03 −0.74 −0.12 −0.31 0.23
Kurtosis −0.34 0.01 −0.42 −0.02 -1.27 −0.56 −0.99 −0.89 −0.79 −0.40 −0.69 −0.84 −0.68 0.01 −0.61 −0.44 −0.70

Mean score variables 1 2 3 4
1. SubSsurf
2. TecSurf 0.01
3. ActDeep −0.21 0.21
4. PraDeep −0.12 0.25 0.29
M 2.43 3.18 2.84 3.27
SD 0.75 0.83 0.79 0.77
Skewness 0.41 −0.24 −0.15 −0.28
Kurtosis −0.24 −0.53 −0.36 −0.02

Appendix 4: means, standard deviations and one-way analyses of variance for the effects of intending and not intending to change major subject on the learning approaches of chemistry minors

Table 8 Means, standard deviations and one-way analyses of variance for the effects of intending and not intending to change major subject on the learning approaches of chemistry minors (N = 358)
Intends to change major (N = 65) Doesn’t intend to change major (N = 293) F(1, 355) p η 2
SubSurf 2.53 0.82 2.47 0.74 0.329 0.567 0.001
TecSurf 3.29 0.82 3.11 0.86 2.343 0.127 0.007
ActDeep 2.65 0.83 2.77 0.79 1.189 0.276 0.003
PraDeep 3.18 0.85 3.23 0.77 0.171 0.679 0.001

Appendix 5: Tukey HSD multiple comparisons of learning approaches vs. persistence

Table 9 (a) Means, standard deviations and one-way analyses of variance for the effects of intending and not intending to change major subject on the learning approaches of chemistry minors (N = 358): SubSurf and TecSurf. (b) Means, standard deviations and one-way analyses of variance for the effects of intending and not intending to change major subject on the learning approaches of chemistry minors (N = 358). ActDeep and PraDeep
Dependent variable (I) Persistence (J) Persistence Mean difference (I–J) Std. error Sig. 95% confidence interval
Lower bound Upper bound
a The mean difference is significant at the 0.05 level.
SubSurf Intended and changed Intended but didn't change −0.36414 0.15584 0.139 −0.7941 0.0658
Didn't intend but changed 0.05938 0.20349 0.998 −0.5021 0.6208
Didn't intend and didn't change −0.11300 0.13908 0.926 −0.4967 0.2707
Dropped out −1.01875a 0.26644 0.002 −1.7539 −0.2836
Intended but didn't change Intended and changed 0.36414 0.15584 0.139 −0.0658 0.7941
Didn't intend but changed 0.42352 0.20502 0.240 −0.1422 0.9892
Didn't intend and didn't change 0.25115 0.14130 0.390 −0.1387 0.6410
Dropped out −0.65461 0.26760 0.109 −1.3930 0.0837
Didn't intend but changed Intended and changed −0.05938 0.20349 0.998 −0.6208 0.5021
Intended but didn't change −0.42352 0.20502 0.240 −0.9892 0.1422
Didn't intend and didn't change −0.17237 0.19259 0.898 −0.7038 0.3590
Dropped out −1.07813a 0.29789 0.004 −1.9000 −0.2562
Didn't intend and didn't change Intended and changed 0.11300 0.13908 0.926 −0.2707 0.4967
Intended but didn't change −0.25115 0.14130 0.390 −0.6410 0.1387
Didn't intend but changed 0.17237 0.19259 0.898 −0.3590 0.7038
Dropped out −0.90575a 0.25820 0.005 −1.6182 −0.1933
Dropped out Intended and changed 1.01875a 0.26644 0.002 0.2836 1.7539
Intended but didn't change 0.65461 0.26760 0.109 −0.0837 1.3930
Didn't intend but changed 1.07813a 0.29789 0.004 0.2562 1.9000
Didn't intend and didn't change 0.90575a 0.25820 0.005 0.1933 1.6182
TecSurf Intended and changed Intended but didn't change 0.03237 0.18771 1.000 −0.4855 0.5503
Didn't intend but changed −0.20250 0.24511 0.922 −0.8788 0.4738
Didn't intend and didn't change 0.14750 0.16702 0.903 −0.3133 0.6083
Dropped out 0.31000 0.32093 0.870 −0.5754 1.1954
Intended but didn't change Intended and changed −0.03237 0.18771 1.000 −0.5503 0.4855
Didn't intend but changed −0.23487 0.24695 0.876 −0.9162 0.4464
Didn't intend and didn't change 0.11513 0.16970 0.961 −0.3531 0.5833
Dropped out 0.27763 0.32233 0.911 −0.6117 1.1669
Didn't intend but changed Intended and changed 0.20250 0.24511 0.922 −0.4738 0.8788
Intended but didn't change 0.23487 0.24695 0.876 −0.4464 0.9162
Didn't intend and didn't change 0.35000 0.23161 0.557 −0.2890 0.9890
Dropped out 0.51250 0.35881 0.610 −0.4774 1.5024
Didn't intend and didn't change Intended and changed −0.14750 0.16702 0.903 −0.6083 0.3133
Intended but didn't change −0.11513 0.16970 0.961 −0.5833 0.3531
Didn't intend but changed −0.35000 0.23161 0.557 −0.9890 0.2890
Dropped out 0.16250 0.31074 0.985 −0.6948 1.0198
Dropped out Intended and changed −0.31000 0.32093 0.870 −1.1954 0.5754
Intended but didn't change −0.27763 0.32233 0.911 −1.1669 0.6117
Didn't intend but changed −0.51250 0.35881 0.610 −1.5024 0.4774
Didn't intend and didn't change −0.16250 0.31074 0.985 −1.0198 0.6948
ActDeep Intended and changed Intended but didn't change −0.00828 0.17918 1.000 −0.5027 0.4862
Didn't intend but changed −0.48125 0.23238 0.238 −1.1225 0.1600
Didn't intend and didn't change −0.21538 0.15882 0.657 −0.6536 0.2229
Dropped out 0.36250 0.30425 0.756 −0.4770 1.2020
Intended but didn't change Intended and changed 0.00828 0.17918 1.000 −0.4862 0.5027
Didn't intend but changed −0.47297 0.23505 0.265 −1.1216 0.1756
Didn't intend and didn't change −0.20710 0.16271 0.708 −0.6561 0.2419
Dropped out 0.37078 0.30630 0.745 −0.4744 1.2160
Didn't intend but changed Intended and changed 0.48125 0.23238 0.238 −0.1600 1.1225
Intended but didn't change 0.47297 0.23505 0.265 −0.1756 1.1216
Didn't intend and didn't change 0.26587 0.21992 0.746 −0.3410 0.8727
Dropped out 0.84375 0.34016 0.100 −0.0949 1.7824
Didn't intend and didn't change Intended and changed 0.21538 0.15882 0.657 −0.2229 0.6536
Intended but didn't change 0.20710 0.16271 0.708 −0.2419 0.6561
Didn't intend but changed −0.26587 0.21992 0.746 −0.8727 0.3410
Dropped out 0.57788 0.29485 0.291 −0.2357 1.3915
Dropped out Intended and changed −0.36250 0.30425 0.756 −1.2020 0.4770
Intended but didn't change −0.37078 0.30630 0.745 −1.2160 0.4744
Didn't intend but changed −0.84375 0.34016 0.100 −1.7824 0.0949
Didn't intend and didn't change −0.57788 0.29485 0.291 −1.3915 0.2357
PraDeep Intended and changed Intended but didn't change −0.18717 0.15840 0.762 −0.6242 0.2499
Didn't intend but changed −0.25625 0.20683 0.729 −0.8269 0.3144
Didn't intend and didn't change −0.55313a 0.14093 0.001 −0.9420 −0.1643
Dropped out −0.49732 0.28648 0.415 −1.2877 0.2931
Intended but didn't change Intended and changed 0.18717 0.15840 0.762 −0.2499 0.6242
Didn't intend but changed −0.06908 0.20838 0.997 −0.6440 0.5059
Didn't intend and didn't change −0.36595 0.14320 0.084 −0.7611 0.0291
Dropped out −0.31015 0.28760 0.817 −1.1037 0.4834
Didn't intend but changed Intended and changed 0.25625 0.20683 0.729 −0.3144 0.8269
Intended but didn't change 0.06908 0.20838 0.997 −0.5059 0.6440
Didn't intend and didn't change −0.29688 0.19544 0.552 −0.8361 0.2424
Dropped out −0.24107 0.31686 0.941 −1.1153 0.6332
Didn't intend and didn't change Intended and changed 0.55313a 0.14093 0.001 0.1643 0.9420
Intended but didn't change 0.36595 0.14320 0.084 −0.0291 0.7611
Didn't intend but changed 0.29688 0.19544 0.552 −0.2424 0.8361
Dropped out 0.05580 0.27836 1.000 −0.7122 0.8238
Dropped out Intended and changed 0.49732 0.28648 0.415 −0.2931 1.2877
Intended but didn't change 0.31015 0.28760 0.817 −0.4834 1.1037
Didn't intend but changed 0.24107 0.31686 0.941 −0.6332 1.1153
Didn't intend and didn't change −0.05580 0.27836 1.000 −0.8238 0.7122

Appendix 6: nonparametric Independent samples Kruskal–Wallis H test results of learning approaches vs. persistence for the dataset containing chemistry major bachelor students from one university only

Table 10 Nonparametric Independent samples Kruskal–Wallis H test results of learning approaches vs. persistence for the dataset containing chemistry major bachelor students from one university only (N = 177)
Null hypothesis Sig. Decision
The distribution of SubSurf is the same across categories of Persistence 0.001 Reject the hypothesis
The distribution of TecSurf is the same across categories of Persistence 0.586 Retain the null hypothesis
The distribution of ActDeep is the same across categories of Persistence 0.065 Retain the null hypothesis
The distribution of PraDeep is the same across categories of Persistence 0.003 Reject the hypothesis

Appendix 7: Games–Howell multiple comparisons of first year chemistry grades vs. persistence

Table 11 Games–Howell multiple comparisons of first year chemistry grades vs. persistence for the dataset containing chemistry major bachelor students from one university only (N = 177)
(I) Persistence (J) Persistence Mean difference (I–J) Std. error Sig. 95% confidence interval
Lower bound Upper bound
a The mean difference is significant at the 0.05 level.
Intended and changed Intended but didn't change 0.79a 0.19 0.001 0.25 1.33
Didn't intend but changed 0.23 0.24 0.877 −0.48 0.95
Didn't intend and didn't change 0.33 0.19 0.423 −0.20 0.86
Dropped out 0.60 0.47 0.717 −1.02 2.22
Intended but didn't change Intended and changed −0.79a 0.19 0.001 −1.33 −0.25
Didn't intend but changed −0.56 0.24 0.180 −1.27 0.16
Didn't intend and didn't change −0.046 0.19 0.117 −0.99 0.07
Dropped out −0.19 0.47 0.994 −1.81 1.43
Didn't intend but changed Intended and changed −0.23 0.24 0.877 −0.95 0.48
Intended but didn't change 0.56 0.24 0.180 −0.16 1.27
Didn't intend and didn't change 0.10 0.24 0.994 −0.61 0.80
Dropped out 0.37 0.50 0.941 −1.27 2.01
Didn't intend and didn't change Intended and changed −0.33 0.19 0.423 −0.86 0.20
Intended but didn't change 0.46 0.19 0.117 −0.07 0.99
Didn't intend but changed −0.10 0.24 0.994 −0.80 0.61
Dropped out 0.27 0.47 0.975 −1.35 1.89
Dropped out Intended and changed −0.60 0.47 0.717 −2.22 1.02
Intended but didn't change 0.19 0.47 0.994 −1.43 1.81
Didn't intend but changed −0.37 0.50 0.941 −2.01 1.27
Didn't intend and didn't change −0.27 0.47 0.975 −1.89 1.35

Appendix 8: non-parametric pairwise comparisons for grade vs. persistence

Table 12 Non-parametric pairwise comparisons for grade vs. persistence (N = 177)
Sample 1 – Sample 2 Test statistic Std. error Std. test statistic Sig. Adj. sig.
Each row tests the null hypothesis that the Sample 1 and Sample 2 distributions are the same. Asymptotic significances (2-sided tests) are displayed. The significance level is 0.05. Significance values have been adjusted by the Bonferroni correction for multiple tests.
Intended but didn’t change – dropped out −15.148 18.169 −0.834 0.404 1.000
Intended but didn’t change – didn’t intend and didn’t change −22.008 9.747 −2.258 0.024 0.240
Intended but didn’t change – didn’t intend but changed −27.514 14.662 −1.877 0.061 0.606
Intended but didn’t change – intended and changed 39.011 10.731 3.635 0.000 0.003
Dropped out – didn’t intend and didn’t change 6.859 17.387 0.395 0.693 1.000
Dropped out – didn’t intend but changed 12.366 20.549 0.602 0.547 1.000
Dropped out – intended and changed 23.863 17.957 1.329 0.184 1.000
Didn’t intend and didn’t change – didn’t intend but changed 5.507 13.680 0.403 0.687 1.000
Didn’t intend and didn’t change – intended and changed 17.003 9.345 1.819 0.069 0.688
Didn’t intend but changed – intended and changed 11.496 14.398 0.799 0.425 1.000


The authors gratefully thank the following people for their invaluable help with the data collection: Harri Lönnberg, Helmi Neuvonen, Henri Kivelä, Heidi Korhonen, Tuomas Lönnberg, Veli-Matti Vesterinen (University of Turku), Leena Mattila, Tanja Lahtinen, Saara Kaski, Piia Valto, Rose B. Matilainen (University of Jyväskylä), Leila Alvila, Tapani Pakkanen (University of Eastern Finland, Joensuu), Johanna Kärkkäinen (University of Oulu). Finally, Pirjo Vahviala, Mikael Moussa and Heidi Salmento are gratefully thanked for data handling.


  1. Allen J. and Robbins S. B., (2008), Prediction of College Major Persistence Based on Year Academic Performance, Res. High. Educ., 49, 62–79.
  2. Altonji J. G., (1993), The Demand for and Return to Education When Education Outcomes are Uncertain, J. Labor Econ., 11(1), 48–83.
  3. Araque F., Roldán C. and Salguero A., (2009), Factors influencing university drop out rates, Comput. Educ., 53(3), 563–574.
  4. Arjoon J. A., Xu X. and Lewis J. E., (2013), Understanding the State of the Art for Measurement in Chemistry Education Research: Examining the Psychometric Evidence, J. Chem. Educ., 90, 536–545.
  5. Baeten M., et al., (2010), Using student-centred learning environments to stimulate deep approaches to learning: factors encouraging or discouraging their effectiveness, Educ. Res. Rev., 5(3), 243–260.
  6. Bean J. P., (1980), Dropouts and Turnover: The Synthesis and Test of a Causal Model of Student Attrition, Res. High. Educ., 12(2), 155–187.
  7. Bean J. R., (1982), Conceptual Models of Student Attrition: How Theory Can Help the Institutional Researcher, in Pascarella E. T. (ed.), New Directions for Institutional Research: Studying Student Attrition, San Francisco: Jossey-Bass, pp. 17–33.
  8. Bean J. P., (1985), Interaction Effects Based on Class Level in an Explanatory Model of College Student Dropout Syndrome, Am. Educ. Res. J., 22(1), 35–64.
  9. Bean J. and Eaton S. B., (2002), The Psychology Underlying Successful Retention Practices, J. Coll. Stud. Retent., 3(1), 73–89.
  10. Beattie V., Collins B. and McInnes B., (1997), Deep and surface learning: a simple or simplistic dichotomy? Acc. Educ., 6(1), 1–12, available at:
  11. Bennett R., (2003), Determinants of Undergraduate Student Drop Out Rates in a University Business Studies Department, J. Further High. Educ., 27(2), 123–141.
  12. Brower A. M., (1992), The “Second Half” of Student Integration: The Effects of Life Task Predominance on Student Persistence, J. High. Educ., 63(4), 441–462.
  13. Brown T. A., (2006), Confirmatory Factor Analysis for Applied Research, New York, NY: Guilford Press.
  14. Byrne B. M., (2012), Structural Equation Modeling with Mplus: Basic Concepts, Applications, and Programming, New York, NY: Routledge.
  15. Cabrera A. F., Castañeda M. B., et al., (1992a), The Convergence between Two Theories of College Persistence, J. High. Educ., 63(2), 143–164.
  16. Cabrera A. F., Nora A. and Castañeda M. B., (1992b), The Role of Finances in the Persistence Process: A Structural Model, Res. High. Educ., 33(5), 571–593.
  17. Cohen J., (1988), Statistical power analysis for the behavioral sciences, 2nd edn, Hillsdale, NJ: Erlbaum.
  18. Curran P. J., West S. G. and Finch J. F., (1996), The robustness of test statistics to nonnormality and specification error in confirmatory factor analysis, Psychol. Meth., 1(1), 16–29.
  19. Di Pietro G. and Cutillo A., (2008), Degree flexibility and university drop-out: the Italian experience, Econ. Educ. Rev., 27, 546–555.
  20. Dolmans D. H., Wolfhagen I. H. and Ginns P., (2010), Measuring approaches to learning in a problem based learning context, Int. J. Med. Educ., 1, 55–60, available at:
  21. Donche V., et al., (2013), Differential use of learning strategies in first-year higher education: the impact of personality, academic motivation, and teaching strategies, Br. J. Educ. Psychol., 83(2), 238–251.
  22. Entwistle N. and McCune V., (2004), The conceptual bases of study strategy inventories, Educ. Psychol. Rev., 16(4), 325–345.
  23. Finnish Advisory Board on Research Ethics, (2009), Ethical principles of research in the humanities and social and behavioural sciences and proposals for ethical review, pp. 1–17, available at:
  24. French B. F., Immekus J. C. and Oakes W. C., (2005), An Examination of Indicators of Engineering Students’ Success and Persistence, J. Eng. Educ., 419–425.
  25. Haarala-Muhonen A., et al., (2017), How do the different study profiles of first-year students predict their study success, study progress and the completion of degrees? High. Educ., 74(6), 949–962.
  26. Habley W. R., Bloom J. L. and Robbins S., (2012), Increasing Persistence: Research-Based Strategies for College Student Success, Wiley.
  27. Herzog S., (2005), Measuring Determinants of Student Return vs. Dropout/Stopout vs. Transfer: A First-to-Second Year Analysis of New Freshmen, Res. High. Educ., 46(8), 883–928.
  28. Higher Education Statistics Agency (UK), (2018), Non-continuation summary: UK Performance Indicators 2016/17.
  29. Howie P. and Bagnall R., (2013), A critique of the deep and surface approaches to learning model, Teach. High. Educ., 18(4), 389–400.
  30. Hu L. T. and Bentler P. M., (1999), Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives, Struct. Equ. Model., 6(1), 1–55.
  31. Kline R. B., (2016), Principles and practice of structural equation modeling, 4th edn, New York NY: Guilford Press.
  32. Kuh G. D., et al., (2006), What Matters to Student Success: A Review of the Literature Spearheading a Dialog on Student Success, Commissioned Report for the National Symposium on Postsecondary Student Success Spearheading a Dialog on Student Success, 18(July), p. 156.
  33. Lassibille G. and Gómez L. N., (2008), Why do higher education students drop out? Evidence from Spain, Educ. Econ., 16(1), 89–105.
  34. Lastusaari M. and Murtonen M., (2013), University chemistry students’ learning approaches and willingness to change major, Chem. Educ. Res. Pract., 14(4), 496–506, available at:  10.1039/C3RP00045A.
  35. Lastusaari M., Laakkonen E. and Murtonen M., (2016), ChemApproach: validation of a questionnaire to assess the learning approaches of chemistry students, Chem. Educ. Res. Pract., 17, 723–730, available at:  10.1039/C5RP00216H.
  36. Lehmann W., (2007), “I just didn’t feel like I fit in”: the role of habitus in university dropout decisions, Canadian J. High. Educ., 37(2), 89–110.
  37. Lent R. W., Brown S. D. and Larkin K. C., (1984), Relation of Self-Efficacy Expectations to Academic Achievement and Persistence, J. Counsel. Psychol., 31(3), 356–362.
  38. Leppel K., (2001), The impact of major on college persistence among freshmen, High. Educ., 41, 327–342.
  39. Lifton D., Cohen A. and Schlesinger W., (2007), Utilizing First-Year Curricula Linkage to Improve In-Major Persistence to Graduation: Results from a Four-Year Longitudinal Study, Fall 2000–Spring 2004, J. Coll. Stud. Retent., 9(1), 113–125.
  40. Lindblom-Ylänne S., Parpala A. and Postareff L., (2014), Challenges in analyzing change in students’ approaches to learning, in Gijbels D., et al. (ed.), Learning patterns in higher education: dimensions and research perspectives, London: Routledge, pp. 232–248, available at:
  41. Little T. D., (2010), Longitudinal Structural Equation Modeling, New York, NY: Guilford Press.
  42. Lovatt J., Finlayson O. E. and James P., (2007), Evaluation of student engagement with two learning supports in the teaching of 1st year undergraduate chemistry, Chem. Educ. Res. Pract., 8(4), 390.
  43. MacCallum R. C., Browne M. W. and Sugawara H. M., (1996), Power Analysis and Determination of Sample Size for Covariance Structure Modeling, Psychol. Meth., I(2), 130–149.
  44. Marsh H. W., Balla J. R. and McDonald R. P., (1988), Goodness-of-fit indexes in confirmatory factor analysis: the effect of sample size, Psychol. Bull., 103(3), 391–410, available at:
  45. Montmarquette C., Mahseredjian S. and Houle R., (2001), The determinants of university dropouts: a bivariate probability model with sample selection, Econ. Educ. Rev., 20, 475–484.
  46. Muthén B. O., (2002), Beyond SEM: general latent variable modelling, Behaviormetrika, 29(1), 81–117.
  47. Muthén L. K. and Muthén B. O., (2015), Mplus User's Guide, 7th edn, Los Angeles, CA: Muthén & Muthén.
  48. National Center for Education Statistics (USA), (2017), Digest of Education Statistics, 2017: Table 326.10. Graduation rate from first institution attended for first-time, full-time bachelor's degree- seeking students at 4-year postsecondary institutions, by race/ethnicity, time to completion, sex, control of, available at:
  49. Nykänen S.-T., (2013), Kemialla käydään kääntymässä, Jyväskylän Ylioppilaslehti, p. 4.
  50. Ost B., (2010), The role of peers and grades in determining major persistence in the sciences, Econ. Educ. Rev., 29(6), 923–934, available at:
  51. Parpala A., et al., (2010), Students’ approaches to learning and their experiences of the teaching–learning environment in different disciplines, Br. J. Educ. Psychol., 80(2), 269–282.
  52. Pascarella E. T. and Terenzini P. T., (1980), Predicting Freshman Persistence and Voluntary Dropout Decisions from a Theoretical Model, J. High. Educ., 51(1), 60–75.
  53. Pascarella E. T. and Terenzini P. T., (1983), Predicting Voluntary Freshman Year Persistence/Withdrawal Behavior in a Residential University: A Path Analytic Validation of Tinto’ s Model, J. Educ. Psychol., 75(2), 215–226.
  54. Ruuska M., (2010), Katoamistemppu, Ylioppilaslehti.
  55. Smith L. P., et al., (2007), Pharmacy Students’ Approaches to Learning in an Australian University, Am. J. Pharm. Educ., 71(6), 1–120, available at:
  56. Statistics Finland, (2018), Koulutuksen keskeyttäminen väheni edelleen, Helsinki, available at:
  57. St. John E. P., et al., (2004), What Difference Does a Major Make? The Influence of College Major Field on Persistence by African American and White Students, Res. High. Educ., 45(3), 209–232.
  58. Taber K. S., (2017), The Use of Cronbach's Alpha When Developing and Reporting Research Instruments in Science Education, Res. Sci. Educ., 1–24.
  59. Tinto V., (1975), Dropout from Higher Education: A Theoretical Synthesis of Recent Research, Rev. Educ. Res., 45(1), 89–125.
  60. Tinto V., (2006), Research and Practice of Student Retention: What Next? J. Coll. Stud. Retent., 8(1), 1–19.
  61. Tinto V., (2010), From Theory to Action: Exploring the Institutional Conditions for Student, in Smart J. C., (ed.), Higher Education: Handbook of Theory and Research, Springer, pp. 51–89.
  62. Valadas S. T., et al., (2017), The Mediating Effects of Approaches to Learning on the Academic Success of First-Year College Students, Scand. J. Educ. Res., 61(6), 721–734, available at:
  63. Vermunt J., (1994), Inventory of learning styles in higher education, in The Graduate School of Education, Leiden University.
  64. Vermunt J. D. and Donche V., (2017), A Learning Patterns Perspective on Student Learning in Higher Education: State of the Art and Moving Forward, Educ. Psychol. Rev., 29(2), 269–299.
  65. Vermunt J. D. and Vermetten J., (2004), Patterns in Student Learning: Relationships Between Learning Strategies, Conceptions of Learning, and Learning Orientations, Educ. Psychol. Rev., 16(4), 359–384.
  66. Vossensteyn H., et al., (2015), Drop-out and Completion in Higher Education in Europe – Main Report, European Commission, available at:
  67. Wang J. and Wang X., (2012), Structural equation modeling: applications using Mplus, Chichester: John Wiley & Sons.
  68. Wijtmans M., van Rens L. and van Muijlwijk-Koezen J. E., (2014), Activating Students’ Interest and Participation in Lectures and Practical Courses Using Their Electronic Devices, J. Chem. Educ., 91, 1830–1837.
  69. Willcoxson L., Cotter J. and Joy S., (2011), Studies in Higher Education Beyond the first – year experience: the impact on attrition of student experiences throughout undergraduate degree studies in six diverse universities, Stud. High. Educ., 36(3), 331–352.
  70. Wintre M. G. and Bowers C. D., (2007), Predictors of Persistence to Graduation: Extending a Model and Data on the Transition to University Model, Canadian J. Behav. Sci., 39(3), 220–234.
  71. Zeegers P., (2001), Approaches to learning in science: a longitudinal study, Br. J. Educ. Psychol., 71, 115–132, available at:

This journal is © The Royal Society of Chemistry 2019