Reliability

Inter-rater reliability of the total scale score was high (*r* = 0.83), well above the acceptable minimum of 0.70 [45]. Intra-rater reliability, or rating consistency by the same rater two weeks apart, was also high (*r* = 0.94).

#### Factorial Structure

Confirmatory factor analysis was conducted to investigate whether the unidimensional structure of the original RIFL-P instrument applied to the Portuguese version. Model fit for the one-dimensional CFA model was less than optimal (RMSEA = 0.12; CFI = 0.93; TLI = 0.92; cut-off recommendations for acceptable fit require RMSEA < 0.06; CFI > 0.95; TLI > 0.95) [46]. A likelihood ratio test comparing the CFA model to the saturated model was significant, χ2(44) = 143.55, *p* < 0.001. Investigation of the modification indices showed very high correlations across three pairs of items (3 and 8, 1 and 6, and 8 and 10; modification indices were 31.95, 16.88, and 13.52, respectively). Consequently, the insufficient model fit in the CFA analysis was not due to deficient relationships among the items but rather to potential redundancies among some of them. Once the CFA model was modified to take these correlations into account, the model fit indices were high (RMSEA = 0.05; CFI = 0.97; TLI = 0.96). All 11 items had high standardized factor loadings on a single factor (ranging from 0.61 to 0.90, shown in Table 2). The percentage of variance explained by the latent factor in each item (R2) ranged between 0.37 and 0.81. This unidimensional structure was further confirmed with the internal consistency analysis. The Cronbach's alpha was 0.94, with item-total correlations ranging between 0.61 and 0.88.


**Table 2.** Item loadings for the Responsive Interactions for Learning measure in the Brazilian sample.

*Note:* All standardized factor loadings are significant at the 0.001 level. **R2** = coefficient of determination.

Based on the results of psychometric analyses, composite RIFL-P scores were computed as the mean of the 11 items. The mean Brazilian Portuguese RIFL-P score was 2.62 (SD = 0.81), ranging between 1.09 and 4.82. Comparable scores for the original measure, in a sample of Canadian mother–child dyads, had a mean of 3.24 (SD = 0.70), with a minimum score of 1.18 and a maximum score of 4.91 [15]. The differences between the average RIFL-P scores in the Brazilian Portuguese and Canadian samples was statistically significant, t(437) = 8.36, *p* < 0.001, with a large effect size (Cohen's d = 0.82). RIFL-P composite scores were positively correlated with family socioeconomic status (Spearman's *ρ* = 0.46, *p* < 0.001).

#### Convergent and Discriminant Validity

The convergent validity analyses showed that the adapted RIFL-P was significantly associated with all four Brazilian Portuguese PICCOLO domains (*r* = 0.32 for affection; *r* = 0.37 for responsiveness; *r* = 0.41 for encouragement; and *r* = 0.47 for teaching, *p* < 0.001) and the PICCOLO total score (*r* = 0.44, *p* < 0.001). In addition, the adapted RIFL-P was significantly associated with the Five Stimulation Markers (*r* = 0.34, *p* < 0.01) and a range of child outcomes assessed by the INTER-NDA, including cognition (*r* = 0.29, *p* < 0.001), language (*r* = 0.28, *p* < 0.001), and positive behavior (*r* = 0.17, *p* < 0.05). Its association with fine motor skills (*r* = 0.16, *p* = 0.055) was borderline significant and it was not significantly associated with gross motor skills (*r* = 0.13, *p* = 0.110) or negative behavior scores (*r* = −0.001, *p* = 0.992). With respect to discriminant validity, RIFL-P was not expected to correlate with gender (*r* = −0.22, *p* < 0.01) but it did. The RIFL-P scores for boys were lower than for girls.
