According to Angell (1904), perception is the “consciousness of particular material things present to sense.” Similarly, Lindsay and Norman (1977) defined perception as the process by which people “interpret and organize sensation to produce a meaningful experience of the world” based on their information and understanding. Perception relates to how people perceive and interpret any object. Understanding individuals’ perception is crucial to educational practices, as research has shown that perception can serve as a predictor of learners’ motivation to learn and expectation of learning (Martens et al., 2007) and that it can help deepen the understanding of their learning attitude and knowledge (Maio and Haddock, 2009).

In the field of CT and programming education, several attempts have been made to explore the idea of perception and its relationships with other cognitive and affective factors. To measure teachers’ perception of programming, Kong and Wang (2019, 2020) created a questionnaire scale called Perception of Programming Education, measuring from three dimensions: programming understanding, programming expectation, and programming support. Sands et al. (2018) developed a survey on teachers’ perceptions of CT to explore the activities that teachers perceive to be involved in CT. The CT Scales (Korkmaz et al., 2017) is a validated questionnaire that is occasionally used to measure students’ CT perceptions. Nonetheless, not many studies have conceptualized perception using detailed constructs, and within the limited number of studies on students’ perceptions of CT and programming, most of them examined the concept of perception using binary categories (e.g., “positive and negative” and “like and dislike”) via self-designed questionnaires or interview questions (Zainal et al., 2012; Adler and Kim, 2018; Wong and Cheung, 2020).

Research has shown that a positive perception of learning is correlated with motivation as well as learning performance in various contexts. Miyazoe and Anderson (2010) analyzed students’ perceptions of learning English through questionnaires and noticed that students with higher levels of perceived usefulness of English language learning performed better in English writing tests. Fabian et al. (2018) conducted a correlation test, and the results showed that students’ positive perceptions of mathematics learning activities also lead to higher achievement in mathematics. In programming education, research has revealed the significant influence of perceived programming skills on learning motivation, with pre-course perceptions having no overall influence on students’ performance (Zainal et al., 2012). Few empirical studies have explored the relationships among perceptions, motivation, and performance in CT.

Instead of using a simple binary definition of perception, in this current study, we examine the perceptions of CT from three sub-dimensions: perceived impact, perceived interest, and perceived utility, which are significant factors of perception constructs (Kong and Wang, 2019). The perceived impact of CT refers to the degree to which accomplishing a task is perceived to “make a difference in the scheme of things” (Frymier et al., 1996). Perceived interest in CT refers to the degree to which students show positive feelings toward, greater concentration on, and an enduring predisposition toward CT learning activities (Kong et al., 2018). Perceived utility of CT refers to the degree to which students perceive CT fitting into their future academic and career plans (Wigfield and Cambria, 2010).

Taking various aspects of learning motivation into consideration facilitates the study of complex learning environments (Mayer, 2009). In this study, we considered the concept of motivation from the following four motivational constructs: self-efficacy, learning goal orientation (LGO), performance goal orientation (PGO), and learning value. All of these constructs have been suggested as potential determinants of academic success (Nolen and Haladyna, 1989; Tuan et al., 2005; Schunk and Zimmerman, 2012).

Self-efficacy refers to people’s belief in their capacity to “execute behaviors necessary to produce specific performance attainments” (Bandura, 1986). Considerable empirical research has indicated the significant positive influence of self-efficacy on students’ learning performance in the contexts of mathematics (Schöber et al., 2018), science (Aurah, 2017; Uçar and Sungur, 2017), and language learning (Namaziandost and Çakmak, 2020). However, the findings on the relationship between self-efficacy and learning performance in the context of CT are inconsistent. In a quasi-experimental study, CT education through digital programming story design has been shown to be effective in increasing both learners’ self-efficacy and learning achievement (Durak, 2018; Durak et al., 2019). The effectiveness of promoting CT through robotics activities has also been validated based on a learning intervention study (Jaipal-Jamani and Angeli, 2017). However, empirical evidence from an experimental study indicated a low correlation between CT and self-efficacy for elementary school students (Wei et al., 2021).

Goal orientation is described as one’s goal preferences in learning or achievement contexts (Payne et al., 2007), and it is usually distinguished into LGO and PGO. LGO refers to students’ preference for self-development through the acquisition of new skills, the mastery of new situations, and the improvement of their competencies, whereas PGO refers to students’ preference for peer competition and teacher attention during the learning process (VandeWalle and Cummings, 1997). Goal orientation is not a bipolar construct, with research suggesting that students can simultaneously have high levels of LGO and PGO (Button et al., 1996). Numerous studies have shown that LGO has positive effects on learning performance, whereas studies on PGO have revealed mixed effects. Through the comparison of pre- and post-test results, Gerhardt and Brown (2006) showed that higher levels of LGO result in higher levels of self-efficacy and better performance, whereas higher levels of PGO lead to a decrease in self-efficacy and performance. From the analysis of students with different levels of academic performance, Hsieh et al. (2007) also suggested that LGO is significantly positively related to students’ academic standing, but PGO shows no such relationship.

In the context of programming education, empirical research has proven that higher levels of LGO result in better programming performance (Bergin et al., 2005). Research in the context of a programming massive open online course (Polso et al., 2020) indicated that students with higher levels of combined LGO and PGO perform better than others, albeit slightly. However, Yukselturk and Bulut (2005) found inconsistent results: in their first course, LGO was positively correlated with programming achievement, but PGO was negatively correlated with programming achievement; in the second course, both LGO and PGO were positively correlated with programming achievement. In the context of CT, Gong et al. (2020) showed that overall learning motivation significantly influences CT skills. However, they did not more closely examine goal orientation and CT skills.

The value of developing CT skills for students includes but is not limited to acquiring CT competencies, experiencing computational problem-solving activities, stimulating their thinking, and identifying the connections between CT and daily life (Wing, 2011; Grover and Pea, 2013). Learning value, also interchangeably referred to as task value, is a key motivational construct, and it has shown a significant correlation with expectancy and knowledge learning (Eccles and Wigfield, 1995; Wigfield et al., 2012).

Bong (2001) found learning value to be a predictor of students’ academic performance, whereas Neuville et al. (2007) showed that learning value could only influence students’ future course enrolment intention and was not correlated with academic performance. In the context of CS education, empirical studies have indicated that students with high levels of learning value perform better in programming (Bergin et al., 2005; Román-González et al., 2018). However, some studies have presented contradictory results, showing that learning value has no significant relationship with programming performance (Watson et al., 2014). More recently, Guggemos (2021) examined various predictors of CT for high school students and found that learning value can predict students’ level but not their development of CT skills.

Inconsistent findings of motivational constructs in the CT context call for further investigation. Therefore, we conducted this study from the alternative perspective of students’ perception of CT skills, and we examined the influence of this perception on students’ motivational constructs and performance in the CT acquisition process.

The empirical research on CT education has revealed some gender differences in students’ affective and motivational development. In a game-based CT instruction workshop, Horn et al. (2016) discovered that, for middle school students, males were more interested in topics like modern technology and adventure stories, but females showed more interest in language literature-related topics. Angeli and Valanides (2020) found that males benefited more from kinesthetic and manipulative learning activities while females were more engaged in collaborative writings. In terms of motivation to learn, Sullivan et al. (2015) noticed that through an intervention in a programming club, male students demonstrated significantly higher levels of self-efficacy than female students in CS. In addition, they found that after the intervention, males expressed stronger aspirations to pursue a bachelor’s degree in CS, but females did not report the same intention. More recently, Wei et al.’s (2021) study echoed previous empirical evidence that male students’ self-efficacy in learning CT could be improved significantly through collaborative programming activities, while female students showed no significant effect.

Furthermore, Research has indicated inconsistent results in the development of CT across gender groups. Some studies reported that male and female students had no significant differences in cognitive learning outcomes in CT (Sullivan and Bers, 2013; Buffum et al., 2016; Chou, 2020). However, conflicting results also existed. Dagienë et al. (2015) reported that male students performed better in tasks requiring spatial thinking skills while female students performed better in tasks with clear instructions in CT assessments. Similarly, Kim et al. (2021) revealed that in a CT literacy test, male students scored higher in programming skills, but female students performed better in abstraction skills. More research on the development of CT across genders is required.

Based on the foregoing discussion, we explore the following research questions: (a) What are the effects of high school students’ perceptions of CT on different CT learning motivational factors (i.e., self-efficacy, learning values, LGO, and PGO) and on CT learning performance? (b) What are the effects of different motivational factors on high school students’ CT learning performance? (c) Do the effects stated above differ by gender?

Based on review studies, we hypothesize the causal model presented in Figure 1 to explore the direct effects of the perceptions of CT on motivational factors (i.e., self-efficacy, LGO, PGO, and learning value) and on students’ learning performance in CT. We also examined the potential effect of motivational factors on students’ learning performance in CT. We further examined whether the structural model differs across different gender groups.

We hypothesize that (a) the perceptions of CT positively influence all four motivational factors (i.e., self-efficacy, learning values, LGO, and PGO) and students’ CT performance. We also hypothesize that (b) students’ self-efficacy, LGO, PGO, and learning value positively influence their CT performance. In addition, we hypothesize that (c) there is no significant difference in the structural model coefficients between the male and female student groups.

This study was conducted in an urban high school in China. Through convenience sampling, students from Grades 10 and 11 were invited to voluntarily participate in the study. The student demographic comprised mostly Chinese students (99%) along with students from other races (1%). Of the 285 students recruited, 144 (50.53%) were male, 141 (49.47%) were female, and their average age was 16. The school had implemented a CS curriculum called Information Technology; CT is embedded in the curriculum through learning Python programming. The participating students had taken the programming course once per week for 3 months and had a basic knowledge of Python programming and CT concepts. For the data collection, two questionnaires on perceptions and learning motivation of CT and one performance test of CT were conducted. All questionnaires and test items used in the experiment were in Chinese in consideration of students’ language preferences. Participants were asked to indicate their gender (1 = male, 2 = female) as demographic information at the beginning of the questionnaires.

Kline (2016) proposed that sample sizes over 200 can be considered large for a structural equation modeling (SEM) analysis. Bentler and Chou (1987) also suggested a sample size to estimated parameters (N:p) ratio of 5:1 as acceptable for a structural equation model. The 46 free parameters and 285 participants in this study yielded a ratio of 6.2:1. Therefore, based on the criteria, the sample size in this study was sufficient to address the hypotheses.

IBM SPSS Amos 27 software (Arbuckle, 2020) was used for the SEM analysis. Maximum likelihood estimation was used in the analysis. A confirmatory factor analysis was conducted for the measurement model for the five latent variables (perceptions of CT, self-efficacy, LGO, PGO, and CT learning value). Cronbach’s alpha, the composite reliability (CR), and the average variance extracted (AVE) were reported. Cronbach’s alpha and CR are commonly used methods to assess the reliability of questionnaires. CR performs better than Cronbach’s alpha in questionnaires with multidimensional scales (McNeish, 2018). AVE is usually used to measure discriminant validity in SEM analysis (Fornell and Larcker, 1981), which helps explain how much of the variations in the item can be explained by the construct. Given that Cronbach’s alpha coefficients require very strict assumptions and perform better for unidimensional scales (McNeish, 2018), we further presented alternative indicators, CR and AVE, for the reliability of scales.

The structural model was then tested, and the Chi-square test results and other fit indices, namely the comparative fit index (CFI), the goodness-of-fit index (GFI), the adjusted GFI (AGFI), and the root mean square error of approximation (RMSEA), were reported. The Chi-square test is a commonly used global fit index in SEM (Kline, 2016). Besides the Chi-square test, model fit indices mentioned above are also indicators for model evaluation. Specifically, CFI compares the fit of a hypothesized model with the fit of a baseline model; GFI and AGFI represent the proportion of variance accounted for by estimated covariance; RMSEA indicates how far the hypothesized model is from the perfect model (Kline, 2016). Noted that researchers have revealed that Chi-square is sensitive to sample size, and “a Chi-square will almost always be significant (indicating a poor fit) even with only modest sample sizes” (Iacobucci, 2010). It is suggested that if the statistic adjusted by its degrees of freedom was less than 3 (i.e., χ²/df ≤ 3), the model indicates a reasonable fit (Kline, 2016). In the path analysis of the structural model, we reported the path coefficients (i.e., β weights), which indicated the direct effect of one variable on the other variable in the model (Kline, 2016).

Next, measurement invariance and structural invariance analyses were conducted for the multigroup comparison. Different levels of measurement invariance were reported as a prerequisite of the multigroup comparison, including the configural invariance (i.e., free estimation of all parameters across groups without constraint), the metric invariance (i.e., free estimation of parameters across groups except for factor loadings), and the scalar invariance (i.e., free estimation of parameters across groups except for factor loadings and item intercepts). The Satorra–Bentler scaled Chi-square difference was reported (Satorra and Bentler, 2010), and other criteria [ΔCFI, ΔTLI (Tucker-Lewis Index), and ΔRMSEA] were also included (Little, 1997; Cheung and Rensvold, 2002; Chen, 2007). A structural invariance test was then performed. A procedure flowchart of the research is presented in Figure 2.

Table 1 presents the descriptive statistics of the study variables, namely the sample size, mean, standard deviation, skewness, and kurtosis. The skewness values were between −3 and +3, and the kurtosis values were between −10 and +10, suggesting that the data followed the normal distribution (Kline, 2016).

By intercorrelating all latent variables (i.e., perceptions of CT, self-efficacy, LGO, PGO, and CT learning value), a confirmatory factor analysis was performed in the overall sample. Although the Chi-square result was significant, the measurement model indices were within the recommended criteria and indicated a good fit to the data (χ² = 140.25, df = 80, p < 0.001, χ²/df = 1.753, GFI = 0.938, AGFI = 0.907, CFI = 0.987, RMSEA = 0.051).

Table 2 shows the standardized factor loadings of each item as well as the Cronbach’s alpha, the CR, and the AVE for each construct. Cronbach’s alphas higher than 0.90 and CR values greater than 0.60 suggest that the measurement model has good reliability (Bagozzi and Yi, 1988). Standardized factor loadings greater than 0.70 and AVE values greater than 0.50 suggest that the measurement model has good convergent validity (Bagozzi and Yi, 1988; Hair et al., 2006). Table 3 shows that the measurement model has good discriminant validity, as the square root of the AVE for each construct is greater than its correlation with other constructs (Fornell and Larcker, 1981).

Although the Chi-square result was significant, the SEM results (Figure 3) indicated a good fit of the hypothesized model (χ² = 149.61, df = 90, p < 0.001, χ²/df = 1.662, GFI = 0.938, AGFI = 0.906, CFI = 0.987, RMSEA = 0.048). The structural model showed that the students’ perceptions of CT directly and positively influenced their CT performance (β = 0.13, p < 0.05). The students’ perception of CT also positively influenced all four motivational constructs: self-efficacy (β = 0.21, p < 0.01), PGO (β = 0.33, p < 0.01), LGO (β = 0.16, p < 0.05), and CT learning value (β = 0.38, p < 0.01). As for the relationships between the motivational constructs and the students’ CT performance, the path coefficients showed that the students’ CT performance was positively influenced by self-efficacy (β = 0.30, p < 0.01) and LGO (β = 0.17, p < 0.05).

The measurement invariance of the latent variables across gender was tested (Table 4). The configural model demonstrated good fit (χ² = 267.65, df = 160, p < 0.001, GFI = 0.890, AGFI = 0.836, CFI = 0.977, RMSEA = 0.049), suggesting that configural invariance was attained. The model fit indices of the metric and scalar invariance model also showed adequate model fit. For the metric invariance, the increase in χ² was not significant (Δχ² = 13.95, Δdf = 10, p = 0.175) and the changes in the model indices (ΔCFI = −0.001, ΔTLI = −0.001, ΔRMSEA = −0.001) were within the recommended criteria (Cheung and Rensvold, 2002; Chen, 2007), supporting the metric (weak) invariance. For the scalar invariance, although the increase in χ² was significant (Δχ² = 70.15, Δdf = 15, p < 0.001), the changes in the model indices (ΔCFI = −0.01, ΔTLI = −0.01, ΔRMSEA = 0.008) were within the recommended criteria, thus supporting the scalar (strong) invariance of the measurement model. We further tested the residual invariance. The increase in χ² was significant, and the changes in the model indices were beyond the recommended criteria, thus rejecting the residual invariance.

The structural invariance test was conducted as follows. First, the model in which factor loading values were constrained and other parameters were freely estimated across gender groups was estimated as the baseline model; it demonstrated good fit (χ² = 313.75, df = 195, p < 0.001, GFI = 0.880, AGFI = 0.833, CFI = 0.974, RMSEA = 0.046). Second, we tested a fully constrained model in which all of the structural path coefficients were constrained. The comparison results between the fully constrained model and the baseline model (Δχ² = 24.43, Δdf = 4, p < 0.001) indicated significant differences in certain structural path coefficients across gender groups. Third, the critical ratio test was performed to locate the origin of the structural invariance, and the pairwise parameter values were calculated. A critical ratio statistic greater than +1.96 or smaller than −1.96 indicates differences across genders at the 0.05 significance level. Two path coefficients (perceptions of CT self-efficacy and perceptions of CT LGO) differed significantly between the male and female students (see Figure 4). The male students who perceived CT more positively showed higher levels of self-efficacy (β = 0.41, p < 0.001) and LGO (β = 0.27, p < 0.001), whereas the female students’ perception of CT did not have a significant influence on their level of self-efficacy (β = −0.13, p = 0.136) or LGO (β = −0.01, p = 0.938).

In this study, students’ perceptions of CT covered three aspects (i.e., perceived impact, perceived interest, and perceived utility). Students who perceive higher levels of these aspects were found to perceive CT as having a greater impact on society, as being more interesting and attractive, and as being more useful to their academic and career development. The students’ perceptions of CT directly influenced their performance on the CT test. This finding is consistent with the findings of Miyazoe and Anderson (2010) and Fabian et al. (2018).

The results also suggest that students with more positive perceptions of CT tend to report higher levels of motivation for CT acquisition. Research has revealed a significant correlation between one’s perception and motivation in different learning contexts (Chen and Hoshower, 2003; Martens et al., 2007). This study contributes to the field by validating this relationship and by further examining the concept of motivation from a multidimensional perspective: students’ perceptions of CT positively influence their self-efficacy, learning value, and goal orientation.

With the implementation of CT in K-12 curricula worldwide, efforts have been made to develop instructional techniques and strategies. This study’s empirical results highlight the need for more research attention on students’ perceptions of CT to further enrich their affective and cognitive learning outcomes, improve the effectiveness of CT acquisition, and facilitate the development of CT education. By examining students’ perceptions from various dimensions, the results yield insights into the development of CT instructional strategies. Furthermore, they imply that more instructional efforts can be made to improve students’ awareness of how CT affects and changes the way people live in modern society, organize attractive and interesting CT learning activities, and engage students in meaningful activities that are relevant to them.

In this work, self-efficacy was found to exert a significant effect on CT learning performance, echoing the results of Jaipal-Jamani and Angeli (2017) and Durak (2018). Self-efficacy, as the foundation of human motivation (Bandura, 1986), has been recognized as an effective tool for predicting students’ learning performance in the field of CT. This relationship can be explained by the fact that students who experience CT education through programming constantly encounter problems related to their daily life in programming learning, which leads to greater efforts in their academic study, particularly in their CT course.

In terms of goal orientation, we examined the construct using a two-dimension classification: LGO and PGO. LGO positively influenced the students’ CT performance, whereas PGO did not. The positive relationship between LGO and learning performance is consistent with many previous studies (Gerhardt and Brown, 2006; Hsieh et al., 2007). Researchers have explained that students with high levels of LGO demonstrate more positive self-evaluation and are highly committed to self-improvement and growth, which are all supportive of long-term educational aspirations and “boost” academic achievement. However, research on PGO has generated mixed results. Some researchers have argued that LGO and PGO may go “hand in hand” and that students with high levels of PGO can also have high academic achievements, as they have also reported high levels of commitment and effort (Yukselturk and Bulut, 2005; Polso et al., 2020). Other studies have indicated a negative relationship between PGO and performance (Gerhardt and Brown, 2006). Students with high levels of PGO may experience high levels of stress or depressive symptoms, resulting in lower academic achievements (Tuominen-Soini et al., 2008). Our results show no significant relationship between PGO and performance, which reflects a mixed situation, to some extent, and high levels of commitment and stress may coexist in a group of students with high levels of PGO, which obscures the correlation with CT performance. To examine the effect of PGO more accurately, some researchers have recommended further classifying PGO into performance-approach goal orientation and performance-avoidance goal orientation (Elliot and Harackiewicz, 1996; Schmidt and Ford, 2003). Adopting performance-approach goals means that students aim to demonstrate competence among their peers, and adopting performance-avoidance goals means that they aim to avoid judgments of incompetence. This approach may yield more accurate results for the effects of PGO on learning performance and thus is a potential research direction for CT education development.

The other motivational construct examined in the study is learning value. No significant influence of learning value on CT learning performance was found. This result aligns with some empirical studies (Watson et al., 2014) but contradicts others (Bergin et al., 2005; Román-González et al., 2018). It is generally assumed that students who regard CT as more essential and valuable for their success put more effort into and try to overcome obstacles during their CT learning process. The non-significant association between learning value and CT performance may be attributed to the characteristics of the participating students. The teacher in this high school has constantly emphasized the value of CS in the classroom, and most high school students in the twenty-first century have placed a high value on developing CT skills as well as other learning activities.

The measurement invariance results indicate that the measurement model had the same meaning across gender groups and ensured meaningful comparison. The results of the multigroup comparisons reveal that perceptions of CT had a significant positive influence on self-efficacy for the male students but had no significant influence on self-efficacy for the female students. Furthermore, perceptions of CT had a significant positive influence on LGO for the male students but had no significant influence on LGO for the female students. The results suggest that male students tend to exhibit more confidence and more intrinsic impetus in improving their CT skills when they hold positive perceptions of it, whereas female students’ confidence and commitment to self-improvement do not vary significantly along with their perceptions of the learning contents. This finding echoed previous empirical evidence in section “Gender difference” that through interventions of CT, male students exhibited a significant increase in self-efficacy and perceived interest in CT skills (Sullivan et al., 2015; Wei et al., 2021). Previous research has indicated the significant influence of perception of programming on students’ intrinsic motivation (Zainal et al., 2012). We argue that the positive influence of perceptions on self-efficacy and LGO is particularly more salient for male students. This finding provides implications for differentiated instruction such that teachers may devise CT learning activities that promote positive perceptions of the learning content to improve male students’ motivation, whereas there is no special need to consider perceptions in instruction for female students.