A total of 510 2nd and 6th grade students from two different public schools (located in province of the state of São Paulo) participated in this stage of the present study. Among the participants, 48.0% were female and 49.1% were male and 2.9% of the students did not answer the question about gender. These participants were aged between 6 and 11 years with mean age of 8.4 years (SD = 1.24).

A total of 427 2nd and 6th grade students participated in this stage of the present study. These students were from public schools of four different Brazilian states, namely Paraná (29.5%), Mato Grosso (18.7%), Amazonas (23.9%), and Bahia (27.9%). Of the 427 participants, 45.4% were male and 54.3% were female, and only one of them did not answer the question about gender. The students were aged between 7 and 13 years with mean age of 9.3 years (SD = 1.20).

Identification questions: participants’ personal Information, such as name, age, gender, and grade in school was collected and included in the instrument (RACF).

Roteiro para Avaliação da Consciência Fonológica (Santos, 1996): a phonological awareness instrument - This is a screening test composed of three sections of items, each with five items and two examples which aim to evaluate difficulties in identifying a phoneme at the beginning, middle, and end of words. Each item consists of one model word and three other alternative words. For example, if the model word given as a stimulus is “copo” (Brazilian Portuguese), the child should select from among the three possible answers provided, the word that begins with the same sound: cobra, sapo, and lata (Brazilian Portuguese). In this case the word “cobra” is correct. An example using words in English would be: if the model word given is “horse”, the child should select from among the following three answers, the word that begins with the same sound: home, snake, and book; in this case the word “home” is correct. In this task, the child receives one point for each correct answer, and the maximum number of points the child can earn is 15 points. The criterion validity of this instrument was previously confirmed when the authors identified that the higher the school grade, the better the results (Suehiro & Santos, 2011).

The subjects of the present study are part of a larger study which aimed to assess the language skills of elementary and middle school students. One of the purposes was to validate the instrument for the various regions of Brazil without exclusion criteria since all students attending elementary and middle school participated in the study.

The present study was approved by the Research Ethics Committee of the Universidade São Francisco, Itatiba, SP, Brazil (São Francisco University, a Franciscan higher education institution in the state of São Paulo), under the Protocol nº 0273.0.142.000-08. The instrument was administered in rooms selected by the school to students whose parents or guardians signed the written informed consent statement. The instrument was individually administered and supervised by undergraduate and graduate (enrolled in masters or doctorate programs) psychology students. These students were properly trained to ensure the instrument would be administered under the same conditions in every school by following specific instructions and guidelines. The total administration time was approximately 15 minutes.

Exploratory factor analysis was carried out using the Statistical Package for the Social Sciences (SPSS). The statistical model was tested using AMOS (Analysis of Moment Structures) 16 and the maximum likelihood estimation method. The model fit was evaluated in a one-factor model and in a three-factor model. In order to identify the model, the metric of latent variables was fixed at mean 0 and standard deviation 1. For the model fit assessment, four fit indices were used, according to the literature recommendations (Byrne, 2001; Schweizer, 2010). The Chi-square index indicates the magnitude of the discrepancy between the observed and expected (fitted) covariance matrices. High chi-square values indicate inadequate fit; however, Chi-square (AMOS lists relative Chi-square as CMIN/DF [Minimum Discrepancy/Degrees of Freedom] - Chi-square/degree of freedom ratio) is affected by the sample size. Therefore, it is recommended to use a Chi-square/df ratio (χ²/df) less than 2, which indicates good fit. The “Comparative Fit Index” (CFI) analyzes the model fit comparing it to the null model (in which all of the variables are uncorrelated). A CFI value greater than 0.95 is indicative of good fit. The Root-Square Error of Approximation (RMSEA) fit index is also a measure of discrepancy, but it penalizes model complexity. Root-Square Error of Approximation values less than 0.05 indicate a good fit. Finally, the Standardized Root Mean Square Residual (SRMR) fit index indicates the standardized mean of the residues (a residue is the difference discrepancies between the observed and expected matrices). Standardized Root Mean Square Residual values less than 0.10 are indicative of good fit.

Initially, the factorability of the correlation matrix of the 15 items of the instrument (RACF) was examined; it was considered to be factorable: Kaiser-Meyer-Olkin (KMO) = 0.82 and Bartlett’s test of sphericity = χ² (105) = 919.06, p < 0.001. Since the goal was to identify possible latent constructs, EFA was carried using the extraction method Principal Axis Factoring (PAF) and Promax rotation (Fabrigar, Wegener, MacCallum, & Strahan, 1999). To determine the number of factors to retain, the “eigenvalues greater than 1.00 rule” and the Scree test were used (Laros, 2004).

The results of the EFA indicated three factors with eigenvalue greater than 1.0. The Scree test, shown in Figure 1, confirmed the EFA solution; the scree plot shows a substantial drop after the third factor (the first 10 eigenvalues were: 3.35, 1.40, 1.06, 0.99, 0.98, 0.90, 0.87, 0.83, 0.76, and 0.73). These results indicate that three factors were extracted, which explained 38.74% of the total variance. Table 1 shows these results.

Figure 1.

Table 1

Note: PhAw: Phonological Awareness.

As can be seen from Table 1, the items 8 and 15 did not load on the same factors for which they were intended in the task, and were therefore excluded from subsequent statistical analysis. As expected, factors 1 and 2 grouped items corresponding to the same group of sounds, beginning and end sounds, respectively. The first factor was composed of 5 items with saturations ranging between 0.399 and 0.700 and eigenvalue of 3.35, explaining 22.32% of the total variance. Factor 2 was composed of four items with saturations between 0.429 and 0.738 and eigenvalue of 1.40, explaining 9.35% of the total variance. Factor 3, composed of middle sound items, had four items, just like factor 2, with saturations ranging from 0.519 to 0.660 and eigenvalue of 1.06, explaining 7.07% of the total variance.

It can be observed that the instrument items did not load on a single factor; it showed that a three-factor solution fits the data the best. The items were grouped according to the sub-skill measured, i.e., the recognition of the beginning, middle, and ending sounds.

The models were tested in order to find the one that best represents the construct evaluated. The initial hypothesis was that the items formed a single factor; however this was not shown in the exploratory factor analysis. Each sound group (beginning, middle, and ending sounds) formed a factor. Thus, two models were analyzed, a one-factor and a three-factor models, aiming to prove the hypothesis tested in the exploratory analysis, i.e., each “sound” in a factor. The fit indices used in the two models are shown in Table 2.

Table 2

Note: RACF: Protocol for Phonological Awareness Task (Roteiro de Avaliação da Consciência Fonológica); CFI: Comparative Fit Index; RMSEA: Root-Square Error of Approximation; SRMR: Standardized Root Mean Square Residual.

As can be seen from Table 2, the three-factor model had the best fit indices (χ²/df = 1.854, CFI = 0.901, RMSEA = 0.045, and SRMR = 0.0477), as proposed in the exploratory factor analysis. The resulting model is shown in Figure 2 below.

Figure 2.

Figure 2 shows that the highest correlation (r = 0.43) was found between factors 1 (beginning sound) and 2 (ending sound), and the lowest correlation (r = 0.24) was found between factors 1 (beginning sound) and 3 (middle sound). This result shows that there is common process related to phonological awareness among the factors, but there is a difference between them which may be related to the difficulty of the item or to other variables that may be involved in the assessment such as working memory.