A total of 1,168 participants were evaluated, distributed across three grade-level groups: students at the end of 2nd grade and the beginning of 3rd grade (n = 380), at the end of 4th grade and the beginning of 5th grade (n = 408), and at the end of 6th grade and the beginning of 1st year of secondary education (n = 380). Twelve public and fifteen private schools from the metropolitan area of Santo Domingo were randomly selected for participation (see Table 1 for sample stratification). The study was approved by the Research Ethics Committee of Universidad Iberoamericana (code CEI2024–24). After the legal guardians of the participants signed the informed consent form and the participants’ assent was verified, their performance was assessed in two sessions of approximately 45 minutes each—the first including tasks of psycholinguistic precursors of reading, and the second assessing reading performance—by personnel specifically trained by the research team.

Psycholinguistic precursors of reading

Phonological Awareness (PA): Assessed through all phonological awareness subtests of the PROLEXIA battery ( Cuetos et al., 2020), including the 3 subtests for children under 7 years of age and the 5 for those over 7. The total score across the 8 tasks was retained. The ordinal alpha coefficients of the subtests range from.76 to.94, according to the authors.

Rapid Automatized Naming (RAN): Tests for naming colors, objects, numbers, and letters were constructed strictly following the gold standard procedures of Denckla & Rudel (1974).

Short-Term Verbal Memory (STVM): Operationalized through a classical digit-span test, using the Auditory Sequential Memory subtest from the Illinois Test of Psycholinguistic Abilities (ITPA) by Kirk et al. (2004). Cronbach’s alpha for the age range in the present study is between.85 and.87.

Alphabetical Knowledge (AK): The Letter Name or Sound subtest from the PROLEC-R test ( Cuetos et al., 2014) was used, retaining the accuracy measure (percentage of correctly named letters).

Variables related to oral language

Vocabulary (VOC): To obtain a measure of receptive (comprehension) vocabulary, the PEABODY test ( Dunn et al., 2006) was used, retaining the total score. For the age range included in the study, the test manual reports Cronbach’s alpha values between.90 and.94.

Grammatical Structure Comprehension (GSC): The CEG test ( Mendoza et al., 2005) was used, with a Cronbach’s alpha of.91; the total score was retained as the measure.

Oral Comprehension (OC): The oral comprehension subtest from the PROLEC-R test ( Cuetos et al., 2014) was employed.

Reading performance variables

Syllable Reading Fluency: Based on the scores obtained in the Syllable Reading subtest of the ECLEC test ( Cubilla-Bonnetier & Sánchez-Vincitore, 2023), the number of syllables correctly read per minute was calculated. For the age range of this study, the subtest shows Cronbach’s alpha values between.88 and.97.

Word Reading Fluency (WR) and Pseudoword Reading Fluency (PWR): The Word Reading and Pseudoword Reading subtests from the PROLEC-R test ( Cuetos et al., 2014) were used, respectively, with a derived calculation of correctly read words and pseudowords per minute. The overall PROLEC-R test has a Cronbach’s alpha of.79.

Text Reading Fluency (TR): Using the Text Comprehension subtest of the PROLEC-R, reading errors from the first three texts (first two for grades 2–3) were recorded, and the average number of correctly read words per minute was calculated. This measure is expected to be higher than WR, given the presence of functional words that children read automatically within a literary text.

Reading Comprehension: Percentage of correct answers to questions about the first two texts (for grades 2–3) or the first three texts (for grades 4–5 and 6–1st year of secondary) from the Text Comprehension subtest of the PROLEC-R.

All socioeconomic variables were collected through a family questionnaire previously used in the country, with a McDonald’s Ω of.79 ( Cubilla-Bonnetier & Sánchez-Vincitore, 2025): Family income, Mother’s educational level, Father’s educational level, Internet connection at home, Number of children’s books at home, Father living with child at home, Number of people at home/number of adults ratio, and Years of pre-primary education (the different possible values of these variables are shown in Table 2).

To describe the differences between public and private school students in the characterization of various socioeconomic variables, the Chi-square test was used when variables were ordinal, analyzing frequency distributions with proportions, and the Student’s t-test was applied when variables were continuous, described by mean and standard deviation.

To achieve the objective of describing performance differences between public and private school systems across the various psycholinguistic, linguistic, and reading performance variables, since the grouping variable had only two categories, either the Student’s t-test or Welch’s t-test was applied as appropriate (the latter when the assumption of equal variances was violated according to the Brown-Forsythe test). The use of one test or the other can be identified by the presence of corrected degrees of freedom.

To achieve the second objective—determining which socioeconomic variables influence performance in reading fluency, vocabulary, and reading comprehension variables, and to what extent—linear regressions were employed. To verify whether belonging to specific socioeconomic groups is associated with a higher risk of low reading performance, the relative risk tool was used, calculated as: Risk of the group exposed to a condition/Risk of the rest of the sample.

Finally, to calculate the size of the educational gap in reading while controlling for the effect of socioeconomic variables, ANCOVAs were conducted (with each of the studied variables as the dependent variable, the educational system as the fixed factor, and the socioeconomic variables as covariates). All analyses were performed using SPSS (v.27.0.1.0) and JASP (v.0.18.3). This study was not preregistered.

As a preliminary step, students from public and private school systems were characterized according to the different socioeconomic variables, verifying the existence of differences between both groups. As shown in Table 2, significant differences were found in all variables related to the socioeconomic factor, both in the proportional differences of ordinal variables (family income, mother’s and father’s educational level, internet connection at home, number of children’s books at home, and father living with the child) and in the mean differences of continuous variables (number of people/adults ratio at home and years of pre-primary education). In all cases, public school students showed lower levels of well-being across the different variables.

Objective 1: Evaluation of the difference in reading performance between both systems (Size of the educational gap in reading).

For each of the variables representing precursor abilities of reading—whether those related to decoding (phonological awareness, RAN, short-term verbal memory, alphabetic knowledge) or purely linguistic variables (vocabulary, grammatical structure comprehension, or oral comprehension)—as well as for the reading performance variables (fluency in reading syllables, words, pseudowords, and texts, and reading comprehension), differences between public and private school students were examined. As shown in Table 3, in the psycholinguistic abilities that serve as precursors to reading (PA, RAN, and STVM), private school students show better performance (it should be noted that RAN variables are measured as total execution time, so higher values indicate poorer performance), with differences often of medium size (except in some grades for RAN abilities less associated with academic stimulation, such as color or object naming). In vocabulary, the advantage of private school students remains of medium size throughout primary school, approaching a large effect size in 4th grade.

Regarding reading abilities, the differences in decoding skills (reading of syllables, words, pseudowords, and texts) remain consistent throughout primary school in favor of private school students, with medium to large effect sizes. In the domain of reading comprehension, the magnitude of the difference appears to decrease over the course of primary education, starting at a medium level and ending at a small one.

Additionally, at the end of 2nd grade and the beginning of 3rd grade, 13.6% of participants from the private school system were unable to read texts (and therefore could not be evaluated on this task), whereas in the public school system, this percentage rose to 39.3%.

Linear regression models showed that the only variable that consistently predicted reading fluency (words correctly read per minute) across all grades was family income. In grades 2–3, paternal educational level and years of pre-primary education were also significant, though with weaker effects ( Table 4). In grades 4–5, maternal educational level emerged alongside family income, with a similar level of influence. The variance in reading fluency explained by the set of socioeconomic variables remained around 19% between grades 2–3 and 4–5 but decreased to 16.1% in grades 6–1st year of secondary education.

The variable vocabulary was selected as representative of the oral language component. When examining the socioeconomic variables influencing vocabulary, family income once again emerged as significant across all grades. In grades 4–5, maternal educational level also became a predictor, with greater influence. In grades 6–1st year of secondary school, the number of children’s books at home and the presence of internet access at home also appeared as predictors, though with weaker effects ( Table 5). The explanatory power of socioeconomic variables for vocabulary increased steadily throughout primary school (from 12.5% in grades 2–3 to 16.9% in grades 6–1st year of secondary school).

Table 6 presents the results of the linear regression with reading comprehension as the dependent variable and socioeconomic variables as independent predictors. A significant model could not be obtained for grades 6–1st year of secondary school. Once again, family income emerged as a predictor in grades 2–3 and 4–5, with maternal educational level and years of pre-primary education added as predictors of similar importance in grades 4–5.

In contrast to vocabulary, the proportion of variance in reading comprehension explained by socioeconomic variables decreases throughout primary school, starting at 14.9% and eventually becoming non-significant in grades 6–1st year of secondary school.

On the other hand, the relative risk (RR) of having low text reading fluency (1st quartile) for students belonging to the lowest income group (28% of the sample) is 2.2—that is, students from low-income families are more than twice as likely to exhibit low reading fluency. Similarly, students whose mothers have a low educational level (primary education completed or not completed) are 2.54 times more likely to have low text reading fluency. In the domain of reading comprehension, the 28% of the sample with low income has a 1.53 times higher risk of poor reading comprehension (1st quartile), while having a mother with a low educational level increases this risk by 1.54 times.

To isolate the effect of the educational system on reading performance from that of socioeconomic conditions—given that the public system serves students from more vulnerable families ( Table 2)—an ANCOVA was conducted for each outcome variable at each grade level to calculate the performance difference between both educational systems while controlling for socioeconomic status ( Table 7). In the psycholinguistic variables that are precursors to reading, the differences between systems tended to decrease considerably (large effects became medium or small, and medium effects became small), if not disappear altogether. Overall, differences in oral language variables (vocabulary, grammatical structure comprehension, and oral comprehension) tended to disappear (except for vocabulary). Regarding reading performance, in grades 2–3, the differences in syllable and word reading became small, while differences in pseudoword and text reading, as well as in reading comprehension, disappeared. In grades 4–5 and 6–1st year of secondary school, although to a lesser extent, the magnitude of the differences decreased substantially (for example, the difference in reading comprehension in grades 6–1st year of secondary school ceased to be significant). However, medium effect sizes tended to persist in the decoding fluency variables.

Several methodological limitations should be acknowledged. First, this study is cross-sectional: comparisons across grade bands reflect different cohorts rather than the same students followed over time. Accordingly, language describing how gaps ‘develop’ or ‘increase’ across grades should be understood as descriptive of cross-sectional patterns rather than as developmental trajectories. Second, the public–private comparison is observational, since it is impossible to randomly assign students to school type. Although ANCOVA was used to adjust for measured SES variables, this approach assumes that the included covariates adequately capture the selection process—a strong assumption that cannot be fully verified. Unmeasured confounders may remain, and the adjusted differences between systems should therefore be interpreted with caution rather than as estimates of causal system effects. Third, the SES variables were entered as separate predictors in the regression models. While this approach offers broad conceptual coverage, multicollinearity among these indicators is plausible and may affect the stability of individual coefficients; results should therefore be interpreted at the level of the overall model rather than individual predictors. Finally, the persistence of medium effect sizes in decoding fluency after SES adjustment is interpreted as a possible residual system effect; however, without direct measures of instructional quality or quantity, this interpretation remains speculative and should be treated as a hypothesis for future research. Regarding generalizability, although a random sample of schools was used, all were located in the metropolitan area of Santo Domingo, which represents approximately 35% of the country’s population. Future research should include representative samples from other regions of the country.

Notwithstanding these limitations, the strength of this study lies in examining a theoretically differentiated set of reading components, and not only a global reading outcome. This allows for a more granular analysis of how different aspects of reading development are differentially sensitive to SES conditions versus schooling context. Prior work has established that SES-related gaps in reading are not uniform across skills ( Hoff, 2013), in the sense that language-based outcomes such as vocabulary tend to be strongly tied to family linguistic environment, while decoding-related skills may reflect instructional exposure more directly. The present study tests whether this differential sensitivity is observable in a low-resource Latin American context where both SES inequality and instructional quality variation are pronounced.

One of the key findings of this study is that students in the public education system in the Dominican Republic begin primary school with less time spent in pre-primary education compared to those in the private system. The years of pre-primary education serve, among other purposes, to foster language development and stimulate the foundational skills required for reading acquisition. It is necessary to continue the ongoing national effort to universalize pre-primary education, as it is essential for building the foundation of future reading learning. Additionally, public early education should strengthen the stimulation of the essential skills underlying decoding and reading comprehension that were found to be underdeveloped in 2nd grade in the present study: phonological awareness, alphabetic knowledge, vocabulary, and grammatical structure comprehension. Although no significant differences were found between systems in the latter two skills once SES was controlled, schools should play a compensatory role in addressing disparities created by students’ contextual backgrounds.

Furthermore, it is essential to address the educational emergency represented by the fact that 2 out of every 5 students in public schools lack functional reading ability at the transition between 2nd and 3rd grade. This situation, in addition to requiring systematic and early detection systems for low decoding levels (before comprehension difficulties become evident), demands the implementation of early and progressive support mechanisms with increasing levels of individualized attention—such as those proposed by the Response to Intervention (RtI) model—which has accumulated extensive evidence regarding its effectiveness ( Burns et al., 2005). It should be noted that these recommendations are grounded in the broader empirical literature on early literacy intervention and are consistent with the patterns observed in the present data; they are not directly derived from the correlational analyses reported here, which do not permit causal inference.