Writing Self-efficacy and the^{Bruning et al. (2013}) Model

Being able to write successfully is a key competence in the current models of education (^{Zumbrunn, Marrs, & Mewborn, 2016}). Writing is a complex cognitive act that could be highly demanding, especially for novice writers (^{Flower & Hayes, 1980}). According with the review by ^{Pajares (2003}), writing is also an emotional task, and self-efficacy for writing makes both direct and indirect contributions to such writing outcomes as essay scores, grade goals, lower writing anxiety, depth of processes, and expected outcomes.

Although self-efficacy beliefs can be seen as a general construct (e.g. Baessler & Schwarzer, 1996), ^{Bandura (1977}) states that efficacy beliefs on specific tasks (being able to read an academic text, or being able to write an essay, etc.) is a better predictor of the performance of these tasks. In other words, as higher is the fit between the efficacy beliefs measured and the behavior analyzed, higher the predictive power of self-efficacy (Bandura, 2006). The latter rationale is the mainstream approach of educational researchers (e.g. ^{Rosario et al., 2011}, 2012), including those focused on writing self-efficacy (e.g. ^{García & de Caso, 2006a}, 2006b; ^{Pajares, Hartley, & Valiante, 2001}; ^{Zimmerman, Bonner, & Kovach, 1996}), as is the case of the present study.

According to ^{Bruning and Kaufmann (2015}), research on writing self-efficacy follows two different approaches: on one hand, studies associating writing self-efficacy with writing outcomes and conditions which are related with differences in self-efficacy like gender or grade and, on the other hand, those which are focused on self-efficacy as outcome, showing the relevance of self-regulatory processes. Although research efforts on this topic are remarkable, Bruning et al. (2013) claims that is necessary a theoretical model to guide further research. That is, writing is a complex process that can be divided into dimensions, and people may have different efficacy beliefs for each of these dimensions.

Following this rationale, ^{Bruning et al. (2013}) propose a model with three dimensions: (1) Ideation, that is, the ability to generate useful ideas to communicate the objectives of the writer; (2) writing Conventions, the ability to use the accepted standards to put the ideas into written language; and (3) Self-regulation of writing, like being able to manage their emotions, or being able to avoid distractions, etc. To verify its model they developed a scale (i.e. the Self-Efficacy for Writing Scale, SEWS) to assess the three dimensions, and validate the scale and its dimensionality with two studies developed in the USA, one with middle school students and the other with high-school students. Continuing with their work, through the present paper we want to validate the SEWS in another kind of students (i.e. university students) and country (Spain).

The Present Study

In Spain, there are not so much instruments to assess self-efficacy for writing, especially validated at the University-level. Until the best of our knowledge, only two scales exists: the Cuestionario de Evaluación de Autoeficacia hacia Procesos Cognitivos Escritores (EAPCE, ^{Álvarez-Fernández & García-Sánchez, 2014}), and the translation of ^{Pajares et al. (2001}) scale performed by ^{Pérez et al. (2008}).

The EAPCE is a 20-item scale focused on cognitive processes that were answered by students between nine and fifteen years old. The items have a Likert 1-7 response format. Unfortunately, the authors do not report the content of the scale, the dimensions included or their psychometric properties excepting reliability. EAPCE is a multidimensional scale with nine dimensions, one for each cognitive process. Based on their papers, where the authors report the significant results (^{Álvarez-Fernández & García-Sánchez, 2014}, 2015), some of the dimensions are thinking about audience, drawing up an outline, and reading text, that can be related with the ideation and self-regulation dimensions of SEWS.

Another scale in Spanish can be found, but is developed in Argentina. It is the translation into Spanish of ^{Pajares et al. (2001}) scale by ^{Pérez et al. (2008}), which has 10 items and is validated with university students and a Likert 0-10 response format. Unlike the EAPCE, this is a unidimensional instrument. Based on their content, it is focused mainly on the conventions dimension of SEWS.

Given the scarceness of appropriate scales (i.e. suitable for Spanish university students) we are interested to fill this gap translating and adapting the SEWS and analyzing its psychometrics properties in a Spanish sample. SEWS scale has a three dimensions (ideation, conventions, and self-regulation) distributed among 16 items in 0-100 response format. Its response format have shown is considered more appropriate and accurate than traditional Likert scales with a small set of responses (^{Bandura, 2006}; ^{Pajares et al., 2001}).

From our point of view, this scale has many advantages such as: (1) it is theory-driven; (2) it assesses three different areas of writing (Ideation, Conventions, and Self-regulation), making easier the further intervention; (3) it is an instrument reliable and valid; (4) it provides a more detailed understanding of dimensions of efficacy that is being activated; and (5) its response format (0-100) have shown better results than the traditional approach with few steps.

Moreover, there is empirical evidence that shows differences among men and women in written expression in academic contexts (e.g. ^{Keller-Margulis, Mercer, Payan, & McGee, 2015}). Additionally, literature also highlights differences in academic self-efficacy related with sex (i.e. ^{Huang, 2013}; ^{Pajares & Valiante, 1999}), although findings have not been consistent (^{García & Fidalgo, 2008}). Thus, we want to explore is, as is expected by prior studies, differences in self-efficacy for writing regarding also exist.

Procedure

An observational, descriptive, longitudinal study was performed. In order to translate the scale, the method used was translation and back-translation of the original instrument by native speakers (^{Hambleton, Merenda, & Spielberger, 2005}). Once translated, the SEWS was included in an online questionnaire that includes sociodemographic information and other two self-efficacy scales. The questionnaire was answered by university students at the beginning of the 1st or 2nd semester, depending on the subject where it was applied.

Participants

Five hundred and twelve university students (78% women, 22% men) from three different Spanish universities were involved in our study. Of them, 43.2% were studying a Degree in Social Work, 19.7% a Degree in Psychology, 14.5% a Degree in Child Education, 10.9% a Degree in Job Relations and Human Resources, 10.5% a Degree in Primary Education, and 1.2% a Master in Educational Orientation. Practically all the participants stated that they need to improve their academic writing skills (97.3%).

A 43.0% of the participants were on its first year of studies, 40.8% on their second year, 14.8% on their third year, and 1.2% were studying a postgraduate. A series of previous ANOVA analyses has been performed to ensure that there are no significant differences on self-efficacy depending on participants’ university degree, to avoid differential selection.

Instruments

Self-Efficacy for Writing Scale (SEWS; ^{Bruning et al., 2013}). It is a scale with 16 items examining three different dimensions of self-efficacy for writing: (1) Ideation (5 items); (2) Conventions (5 items); and (3) Self-regulation (6 items). The items where answered in a Likert scale from 0 (I'm not sure I could do) to 100 (I'm totally sure I could do) according with the recommendations by ^{Bandura (2006}) and ^{Pajares et al. (2001}). The whole scale in English and Spanish can be seen on the appendix.

General Self-Efficacy Scale (Baessler & Schwarzer, 1996). This instrument assesses an overall perception of efficacy. We follow the Spanish adaptation by ^{Sanjuán, Pérez, and Bermúdez (2000}). It has 10 items in a Likert scale from 0 (Totally disagree) to 10 (Totally agree) and its internal consistency is α = .91). A sample item is “I can always manage to solve difficult problems if I try hard enough”.

Self-Efficacy for Writing (^{Pajares et al., 2001}). This is a unidimensional scale focused on perception of efficacy regarding conventions of writing. The scale was translated into Spanish by ^{Pérez et al. (2008}) and validated in Argentina. It has 10 items in a Likert scale from 0 (Totally disagree) to 10 (Totally agree) and its internal consistency is α = .93. A sample item is “Correctly spell all words in a one-page story or composition”.

Data Analysis

The descriptive analyses of the items were made calculating the mean, standard deviation, skewness, kurtosis, minimum and maximum score, and the correlation coefficient between the item and the rest of the scale, as well as the value of Cronbach’s alpha is the item was removed; univariate normality was tested using the Kolmogorov-Smirnov (K-S) test. Also, descriptive statistics (M, SD and Cronbach’s alpha for internal consistency) were computed for the total scores of SEWS, each of its dimensions, the General Self-Efficacy Scale and the Self-Efficacy for Writing. Reliability of the SEWS was estimated using the Spearman-Brown formula (arranging the items by mean in order to compose the two halves). Factor validity was carried out by exploratory factor analysis (EFA). The estimates of Mardia’s multivariate kurtosis and skewness coefficients were high (380.88 and 31.67, respectively, both p < .001), thus rejecting the hypothesis of multivariate normality; as a consequence, an unweighted least squares extraction method was selected. As the dimensions of the original scale are interrelated due the three underlying factors are different components of the writing process, and following the recommendations by ^{Lloret-Segura, Ferreres-Traver, Hernández-Baeza and Tomás-Marco (2014}), the Oblimin rotation method was used. Convergent validity was carried out analyzing the correlations between the SEWS and the remaining self-efficacy scales. Lastly, we explored the mean differences regarding sex. All statistical analyses were performed using the program SPSS 22.

Item Analysis, Internal Consistency and Reliability

As can be seen on Table 1, item means varied between 57.52 (Item 5) and 84.86 (Item 7) and overall item average was 68.21. Regarding item standard deviations, they ranged from 16.10 (Item 7) to 23.12 (Item 12). Talking about the distribution of responses, we can see that all items have at least one participant who scores on the top of the scale, but in eight items the minimum value are higher than zero (i.e. has a value of 10 or 20; items 2, 3, 5, 7, 8, 9, 10, and 15). All items show a negative skewness, but different degrees of kurtosis. For all the items, univariate normality hypothesis is rejected.

Table 1: SEWS Item analysis. 

Assessing internal consistency, Cronbach’s alpha coefficient for the SEWS dimensions ranged between .89 and .90. All these values are adequate according with ^{Nunnally’s recommendations (1981}) for tests in the validation or adaptation stage. As can be seen on Table 1, removing the item 6 increases the Cronbach’s alpha of Ideations dimension from .89 to .90, but giving the low increase in the dimension’s consistency and its contribution to the scale’s overall consistency, we keep the item in our analyses.

EFA

EFA was performed using the correlations between items to statistically determine their underlying factors. The results of the EFA are showed in Table 2. As can be see there, both the KMO test, .93, and Bartlett’s test of sphericity, χ²(120) = 5886.60, p < .001, support the adequacy of the data for use of EFA. Given than item communalities are .50 or higher we consider our sample size as adequate (^{Lloret-Segura et al., 2014}). Regarding the dimensionality of SEWS, and following the K1 method, we retained three dimensions composed by the same items as in the original scale, that account for 65.86% of the items total variance. The first factor, composed by the items 1, 3, 2, 4, and 5 is Ideation, and has an eigenvalue of 7.78 and explains 48.65% of the variance. The second factor is composed by items 8, 9, 7, 6, and 10 is Conventions, and has an eigenvalue of 1.60 and explains 9.97% of the variance. The last factor is composed by items 12, 11, 13, 14, 15, and 16 is Self-regulation, which an eigenvalue of 1.16 and an explained variance of 7.24%.

Table 2: SEWS results from Exploratory Factor Analysis. Structure matrix. 

Scales Scores and Convergent Validity

Some descriptive statistics for the instruments used can be found on Table 3. Continuing with the analysis, we computed the Spearman’ correlations because neither of the scales nor the dimensions of the SEWS fit to the normal distribution, according to K-S tests. The results are on Table 3. As we can see there, all the dimensions of SEWS are linearly and directly correlated. Between dimensions, the higher association is between Ideation and Self-regulation dimensions, as in the original ^{Bruning et al. (2013}). More interesting for our research is the relationship with the other self-efficacy scales. The association of the SEWS dimensions with the ^{Pajares et al.’ (2001}) self-efficacy writing scale is statistically significant (being the highest correlation with Conventions, r = .78, p ≤ .010), as is expected in the measure of the same construct. Regarding the general self-efficacy scale, we also found a significant relationship but lower than with the self-efficacy for writing scale (its values are between r = .32and r = .46, all p ≤ .010), as is expected because general self-efficacy is an overall, more general, construct.

Table 3: SEWS descriptive statistics, reliability, and correlations. 

Sex Differences

Finally, we want to explore the existence of differences regarding sex. Means and standard deviations can be seen on Table 4. We found significant differences on SEWS and general self-efficacy suggesting than male students reports higher efficacy beliefs than female students: SEWS-Ideation, t(212.40) = -6.00, p ≤ .001, r ² = .037; SEWS-Conventions, t(509) = -2.21, p < .050, r ² = .007; General self-efficacy, t(509) = -2.20, p < .050, r ² = .012. However, these differences have a small effect size, due the participant’s sex accounts for 3.7% of SEWS-Ideation, 0.07% of SEWS-Conventions, and 1.2% of General self-efficacy. Moreover, no significant differences were found in SEWS-Self-Regulation or Self-Efficacy for Writing scale.

Table 4: SEWS descriptives and mean differences regarding sex.