Examining Reading Growth Profiles among Children of Diverse Language Backgrounds Using Known and Unknown Approaches

Abstract

Previous research examining the reading achievement of children speaking languages other than English at home has often grouped such children into one broad category at kindergarten entry—English learners (ELs). EL categorization and placement, however, can considerably underestimate the heterogeneity present among language minority children in terms of their language proficiency and needed English skills. The purpose of our study is twofold: (i) to compare the early elementary (K-3) reading growth patterns among four language background groups of children according to parents’ reported home language use and (ii) to examine the association between student family demographics and home learning environments and students’ growth trajectories. We utilized a “known group” analytic approach—multiple-group latent growth analysis, and an “unknown-group” approach—growth mixture analysis to investigate the heterogeneity in multilingual students’ reading growth trajectories. The known groups consisted of four language backgrounds based primarily on parents’ self-reported home language. Results indicated a sizable gap between LEP children and the other language background groups, albeit with the LEP group demonstrating significant greater growth during first grade than the others. The unknown-group approach identified four emergent reading profiles with different kindergarten entry skills and growth rates. We found English bilingual students were overrepresented in the medium-high and high achieving groups. Further, family socioeconomic status and home literacy practices were robust predictors of children’s reading achievement progress throughout the study, a finding diverging from the known-group approach. We discuss ways the study extends previous research examining language minority students.

Notes

1 Following Shanley (2016), who estimated math growth models using the ECLS-K data, we considered missing values as missing at random and estimated the models using full information maximum likelihood with the appropriate child panel weight (C1_5SC0), which adjusted for nonresponse across the first five waves (K-3).

2 Our final model with two growth periods and different growth trajectories across groups fit the data better than the baseline model with a single linear growth period that was the same across groups (Δ chi-square = 1943.096, 15 df, p < .001) and a baseline growth model with a single nonlinear growth period that was different across groups (Δ chi-square = 1827.948,12 df, p < .001).

3 See Muthén and Muthén (2010) and Leroux (2019) for other examples of models with multiple growth periods.

4 We tested whether the initial status, growth rate1 and growth rate2 were the same across groups and rejected each null hypothesis based on likelihood ratio tests (initial status = ΔX² (3 df) = 93.394 p < .001; growth1 = ΔX² (3 df) =22.690, p < .001; ΔX² (3 df) =15.436, p < .005) and changes in common SEM fit indices.

5 We tested the equality of these effects by applying equality constraints to the parameters across each group and noting the change in model deviance (-2LL). The general form of this test is the slopes are the same (H₀) vs. at least one slope is different from the others (H₁).

6 Following the recommendation in the Mplus output, we replicated this model a number of times with different numbers of random starts (with the last one using over 1,000 random initializations) to ensure the best log likelihood value was replicated each time and the BIC was consistent.

7 In preliminary analyses we allowed the functional form of growth during the last period (spring first to spring third) to be freely estimated within each class but found the best fitting model constrained the estimate to be the same across the latent classes (estimated at 2.17), which indicates a slight slowing of growth between spring first and spring third grade compared to the growth between fall first and spring first grade, as shown in Figure 4.

8 We note that although the odds ratio indicates the change in odds for a one-unit change in the predictor (i.e., from 0 to 1, or from 1 to 2), increases in odds ratios are multiplicative rather than additive.