In the present study we aim to contribute to the body of research by investigating the psychometric quality of the teacher version of TASC in the Spanish context. The validated teacher measure of teaching behaviour will be useful for future research investigating the relevance of the three teaching behaviour components for various student outcomes. Given that English is not commonly practiced in Spanish-speaking contexts, the Spanish version of the instrument will be highly beneficial for research and practice involving classroom practices. Moreover, as long as the population of Spanish-speaking contexts is notable, having validated instruments in this language would contribute to providing more insights about teaching and learning in those geographical areas. In addition, these instruments could be used as a diagnostic tool for improving teaching behaviour and schools in Spanish-speaking contexts more generally.

Undoubtedly, this is a first step which can allow future international comparative studies to test differences between countries and diverse conceptions of education. The present study was motivated by previous research indicating that teachers’ behaviours might universally be important for stimulating students’ self-interest in learning, regardless of their cultural contexts (Maulana, Helms-Lorenz, & Van de Grift, 2015) and the fact that even in collectivist societies teachers’ autonomy support remained important (Soenens, Vansteenkiste, & Van Petegem, 2015). This is an interesting point as long as the emphasis given to social values such as conformity, social harmony, respect to the elders, and obedience to teachers’ rules in collectivist societies may introduce important differences in the satisfaction of the already mentioned basic needs (Maulana & Opdenakker, 2014; Maulana, Helms - Lorenz & Van de Grift, 2015; Maulana et al., 2016).

Our long-term aim is to test this in the Spanish teaching context specifically, and Spanish-speaking contexts more generally. Prior to realizing the long-term aim, the construct and predictive validity of the teaching behaviour should be established first. More specifically, the main research questions are:

The current study included a sample of 410 secondary education teachers teaching 7,114 students in Spain. The teachers sample comprised 166 (40.5%) males and 244 (59.5%) females. Only 9% of the teachers had given classes to the selected group of students for more than 3 years. For the majority of the teachers (74.2%) it was their second or third year with the students, while the remaining teachers (16.8%) gave lessons to the group for the first time. Regarding the school type, 46.1% of the teachers taught in general schools, 4.1% in vocational schools, and 49.8% in multitrack schools. The majority of the teachers (69%) worked in public schools whereas 31% taught in private ones. The data involved 62% of language and social science teachers, 20.7% of maths and science teachers, and 17.3% of vocational education and training subjects teachers.

Teachers participated on a voluntary basis. Participants were not compensated for joining the study. Researcher-school agreement was made prior to conducting the survey in schools.

To measure teachers’ perceptions of their own teaching behaviours, the Teacher as Social Context (TASC) Questionnaire (Wellborn et al., 1988) was used. The original questionnaire was translated and back-translated for use in the Spanish context following the international guideline provided by Hambleton, Merenda, and Spielberger (2004).

The questionnaire consisted of 41 items measuring three dimensions: teachers’ involvement (14 items), structure (15 items), and autonomy support (12 items), and provided a 4-point Likert type response format from 1 (strongly disagree) to 4 (strongly agree). The reliability of the original instrument ranged between .83 and .89 (Belmont, Skinner, Wellborn, & Connell, 1992).

To establish the predictive validity of the TASC instrument, the current study included teachers’ report on students’ academic engagement as a criterion. The academic engagement scale developed by Skinner, Kindermann, and Furrer (2009) was used. The scales consisted of 10 items measuring behavioural engagement (5 items) and emotional engagement (5 items) using a 4-response category (1 = completely not true, 4 = completely true). The reliability of the behavioral engagement scale was .82 and .81 in the case of emotional engagement.

Preliminary analysis. Data was firstly explored in order to verify compliance with statistical assumptions, and to detect atypical cases or missing values that could pose bias for the analyses. The percentage of missing cases was between 0.2% and 4.1%. The MCAR test (Little, 1988) was applied to analyse the pattern and type of missing values (X² = 1638.324, df = 1711, p = .894), concluding they were MCAR (missing completely at random). When small MCAR losses (around 5%) are found, any imputation method seems to reasonably replicate the population parameters (Fernández-Alonso, Súarez-Álvarez, & Muniz, 2012), so EM (expectation-maximization) procedure (Dempster, Laird, & Rubin, 1977) was selected.

Main analyses. The questionnaire was then validated according to the basic specifications of the classical test theory (CTT) (Gil Pascual, 2011; Muñiz, 2000) analysing: (1) response behaviour to items (frequency distribution and percentages, items with missing data, removing items with a percentage of missing exceeding 10%, and ceiling and floor effect removing those lower than 15% in the highest and lowest response options, respectively); (2) measures of central tendency and variability; and (3) degree of compatibility of items with the normal curve (skewness and kurtosis), removing the items with skewness below 2 and/or kurtosis below 7 (Curran, West, & Finch, 1996). Before performing these analyses, negatively-worded items were inverted (reverse coded).

Central tendency measures, measures of dispersion and of the shape of distribution are demonstrated in Table 1. High mean values have been obtained for most items (values between 2.185 and 3.505), keeping the standard deviation below 0.8. All the items complied with the criteria of the normal curve, with asymmetry values below 2 and kurtosis values below 7.

To confirm whether the questionnaire was structured in the three factors indicated by the authors of the original questionnaire (Wellborn et al., 1988), a confirmatory factor analysis (CFA) was performed using the maximum likelihood estimation and the covariance matrix of items as input. The 1-factor model was initially tested by combining the 41 elements together, and the fit of the model was compared with the competing three-factor model. Analyses have been done using the AMOS module of the statistical package SPSS 22.0. The goodness-of-fit of the proposed model has been evaluated applying various common fit indexes: p value associated with the chi-square statistic, verifying the null model against the hypothesised model; Standardised Root-Mean-Square Residual (SRMR); Root-Mean-Square Error of Approximation (RMSEA); Goodness-of-Fit Index (GFI); Tucker Lewis Index (TLI); Comparative Fit Index (CFI); Parsimonious Normed Fit Index (PNFI); and Parsimonious Comparative Fit Index (PCFI). The cut-off for CFI, TLI, CFI values was established above .90 for a good fit criteria (Bentler & Bonnet, 1980); PCFI and PNFI values above .50 (James, Mulaik, & Brett, 1982) and RMSEA and SRMR values below .08 for an acceptable fit (MacCallum, Browne, & Sugawara, 1996), or close to .06 for a good fit (Hu & Bentler, 1999; Vanderberg & Lance, 2000). For the chi-square index, the probability of X² > 0.05 indicates a good fit, although this index is very sensitive to sample size. Finally, for the ratio X²/gl, values lower than 2 indicate a good fit (Lévy, Martín, & Román, 2006).

The factorial invariance of the TASC-Spanish teacher measure across genders was analysed using a multi-group confirmatory factor analysis (MGCFA). According to Cheung and Rensvold (2002), measurement invariance testing should satisfy two criteria: a) the measurement model should be adjusted to each group and b) a multigroup analysis should examine the following invariance types: configural invariance (unconstrained model), metric invariance (weak invariance), scalar invariance (strong invariance), and residual invariance (strict invariance).

Evaluation of measurement invariance was based on the examination of the model fit and the interpretation of changes in CFI and RMSEA values. Following the recommendations by Cheung and Rensvold (2002), a ∆RMSEA < .015 and a ∆CFI < .01 from less-constrained (configural) to more-constrained (scalar) models were used as cut-off for measurement invariance between nested models. The robust maximum likelihood chi-square statistic (X²) was also provided, but due to its sensitivity to sample size it cannot be considered highly informative (Wu, Li, & Zumbo, 2007). Correlational analyses were also performed.

Finally, to investigate the average level of perceived involvement, structure, and autonomy support, descriptive analyses were conducted. Proportions of teachers in terms of perceived teaching behaviour in the category low (-1SD), moderate (-1SD < M < +1SD), and good (+1SD) were estimated; c² was run out to analyse relationships between teachers’ gender and each of the categories.

Results of the confirmatory factor analysis using the 41 items showed that the 1-factor model could not be supported, as indicated by the following fit indexes: X²(779) = 3832.814, p < .000, X²/gl = 4.920, RMSEA = .098, SRMR = .0975, GFI = .630, TLI = .442, CFI = .470, PCFI = .447, PNFI = .397. Similarly, the competing 3-factor model indicated a poor fit: X²(776) = 3570.768, p < .000, X²/gl = 4.602, RMSEA = .094, SRMR =. 0944, GFI = .638, TLI = .488, CFI = .515 PCFI = .488, PNFI = .433. Several standard errors interactions obtained from modification indexes were considered and new analyses were carried out to obtain a better fit in both models (Table 2).

The 3-factor model was retained as long as it showed better fit than the 1-factor solution. Nevertheless, the degree of adjustment obtained could not be considered perfect. In fact, the analysis of the factorial weights made it advisable to suppress those items that did not reach at least a weight equal to .30 in any of the factors (Table 3).

Based on the inspection of factor loadings, we found a number of items below the common cut-off of .30. This suggested that some items were not sufficiently contributing to measure the intended construct. Subsequently, the model was run out again deleting these items (starting with those which showed the lowest factor loadings). The excluded items were: items 1, 4, 13, and 14 (involvement), items 16 and 21 (structure), and items 33, 34, 35, 36, 37, and 38 (autonomy).

Finally, with a questionnaire consisting of the remaining 29 items, the fit indexes were: X² = 624.317 (341), p < .000, X²/gl = 1.831, RMSEA = .045, SRMR = .0572, GFI = .900, TLI = .914, CFI = .927, PCFI = .779, PNFI = .718. Therefore, the three dimensions of teaching behaviour were supported satisfactorily.

The contrast of the factorial equivalence test was initially performed with a preliminary analysis. The goodness of fit of the TASC-Spanish teacher measure structure in males and females samples was examined separately using the 29 items set. The fit indexes were adequate and similar to those of the general model. For males, the fit indexes were X² = 509.364 (341), p < .000, X²/gl = 1.494, RMSEA = .055, SRMR = .0668, GFI = .829, TLI = .895, CFI = .912, PCFI = .766, PNFI = .655; for females, the indexes were X² = 555.551 (341), p < .000, X²/gl = 1.629, RMSEA = .051, SRMR = .0658, GFI = .863, TLI = .883, CFI = .902, PCFI = .757, PNFI = .660. Hence, we can conclude that the three factor model showed a satisfactory fit both for male and female samples.

Multisampling analyses were also performed, creating new nested models (Table 4). First, the configural invariance (M1) was examined (the same form without any restriction). The results showed adequate fit indexes (RMSEA = .037, CFI = .906), indicating that the factorial structure of teaching behaviour remained invariant in the two compared groups. This model was considered as a reference for the subsequent nesting of restrictions. The following three hypotheses enunciated three new models, each of which was nested in the previous one. Model 2 (M2) proposed equivalence in the matrix of factor loadings (metric invariance); the model showed adequate fit indexes (RMSEA = .037, CFI = .904), showing very similar fit indexes to the ones obtained in M1 (lower difference between fit indexes: ΔRMSEA < .015, ∆CFI < .01), indicating that there were no differences between the baseline model (M1) and the constraint model (M2); therefore, there were no differences between the factor loadings of the two evaluated samples (male and female). Model 3 (M3) tested the equivalence between intercepts (scalar invariance); the fit indexes showed an acceptable fit (RMSEA = .037, CFI = .900); when comparing the indexes with those obtained in the base model (M1), the difference between the CFI and RMSEA values did not exceed the criterion value; therefore, the invariance of intercepts was supported. In model 4 (M4), the variances and covariances of the error were restricted to be equal between the groups (residual invariance). Because the fit indexes of the model were not adequate (ΔRMSEA was lower than .015, but ΔCFI was -.023), the invariance of the residuals could not be fully supported.

The global reliability of the 29 items was α = .90, indicating good internal consistency. For each separate dimension, the reliability values were .78 (involvement), .82 (structure), and .77 (autonomy support), indicating that the three dimensions were sufficiently reliable.

Correlational analyses between the three TASC scales and behavioral and emotional engagement were run (Table 5). In general, the three domains correlated slightly stronger with emotional engagement (r= .42-.54) than with behavioral engagement (r = .39-.51). Although small to moderate in magnitude, involvement seemed to be the domain with the strongest relationship with both behavioral and emotional engagement. In all cases, correlations showed higher values with students’ emotional engagement than with behavioural one.

The majority of teachers perceived themselves as moderately good in involvement, structure, and autonomy support (Table 5). This fact means that the teachers who formed this sample have developed a positive image of their own teaching activity. It is outstanding that more than 20% of teachers rated themselves as good in providing structure and autonomy. On the other hand, there is an important number of teachers who perceived themselves in a low level of involvement provision. It is also this dimension the one that showed the lowest rating in the good category as perceived by teachers.

No significant statistical differences were obtained when considering gender in any of the domains: involvement (c² = 3.594, p = .166), structure (c² = 0.851, p = .653), and autonomy (c² = 1.686, p = .430). Regarding involvement, it should be noted that the percentage of teachers who had a good perception of their own activity was higher in women than in men (Table 6). A similar tendency could be observed in structure although differences between male and female teachers were slighter. Finally, better ratings were obtained in males than in females when the focus was on good perceived level of autonomy.

Some limitations should be taken into consideration when interpreting these findings. First, the sample was formed by teachers from a specific geographic and cultural context. Caution is needed when trying to extend the findings to other contexts or teachers’ characteristics because of the restricted diversity within the sample regarding ethnicity. Future replication studies in this line should specifically aim to include more geographically diverse Spanish samples. This wider research may also allow a complete explanation concerning the removed items and why they do not contribute to measuring a certain scale in Spain.

Second, teachers participated in this study on a voluntary basis. Future research should attempt to increase the number of participants and randomly sample them. It would also be important to analyse differences between teachers of diverse subjects or who are giving their classes in different educational levels.

Third, teachers self-reported their behaviours inside the classrooms, what makes data susceptible to desirability biases. A multi-informant approach with alternative measures of teacher behaviour (e.g., students’ opinion, external observations) should be used in the future to establish the extent of potential bias in teachers’ reports.

Fourth, the cross-sectional design of the current study did not allow investigation of causal relationships between psychological need satisfaction and student engagement. Therefore, future research using experimental or longitudinal designs is needed to examine the causal and long-term effects. This perspective may allow the exploration of the dynamic nature of teachers’ provision of relatedness, competence, and autonomy support, repeating the measures and developing a system to follow the evolution of teachers and groups of students.

Finally, the situations involved in the current research are mainly focused on the context of secondary education, which raises questions about the generalizability of the results to other age groups. Future studies need to investigate the validity of the Spanish version of TASC in other educational levels, thereby making the necessary adjustments, so that the items selected optimally match up with the internal diversity of each of these stages.