The Impact of Character Complexity on Chinese Literacy: A Generalized Additive Modeling Approach

ABSTRACT

Purpose

We examined the effect of character complexity on early Chinese literacy (word reading and writing). We also investigated whether cognitive skills (phonological awareness, morphological awareness, and rapid automatized naming [RAN]) could moderate the influence of character complexity on literacy outcomes.

Method

Our pre-registered study included a sample of 342 Cantonese – speaking Hong Kong Chinese children (162 males, 180 females, mean age = 7.32, SD = 0.87) from Grade 1 to Grade 3. We used a generalized additive mixed model (GAMM) to estimate the main effect of complexity on word recognition and dictation as well as interaction effects with each of the three cognitive skills. Age and nonverbal IQ were also included as controls.

Results

Character complexity had a slightly curvilinear but consistently negative association with word reading, with odds ratios ranging from .62 to .71 per 10% increase in complexity. RAN and morphological awareness buffered the negative influence of character complexity. For writing to dictation, character complexity was more curvilinear, but character complexity and the interactions between cognitive skills and character complexity were not significant.

Conclusion

Complex visual input can be a barrier to early reading development in Chinese, and fostering morphological awareness may help mitigate this effect.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/10888438.2023.2217967.

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Notes

1. The original sample we used was a smaller sample of a much larger longitudinal dataset. Details about the selection of this subset are documented in the Supplementary Material II.

2. Estimation of omega in R requires a confirmatory factor loading structure using lavaan (Rosseel, 2014) and the psych package (Revelle & Revelle, 2015), but the initial model variance-covariance matrix was not positive-definite, which may be due to the number of binary items that were estimated. To alleviate this issue, we used the “cor.smooth” function from the psych package, which is designed for transforming matrices that are not positive-definite. Results from the follow-up lavaan summary showed no problems with model estimation thereafter.

3. The writing task (Tong et al., 2009) originally included 20 items, but this test later utilized a 25-item version which was made to be more difficult for older participants. We only used the 20-item version of the test, as those who were only tested on the 20-item test would naturally hit a different ceiling of scores compared to those who were tested on all items.

4. The odds ratios and confidence intervals for GAMs and GAMMs vary based on the location of the fitted line. The odds ratios listed are based on 10% incremental increases in raw predictor units and thus fluctuate based on where the percentage “slice” occurs. This results in several odds ratio estimations and their respective confidence intervals for each GAMM model. A detailed list of each odds ratio and its respective confidence interval per percentage increase can be seen in the Supplementary Material V.

Additional information

Funding

This research was supported by the Collaborative Research Fund (CUHK8/CRF/13 G; C4054-17WF to C. McBride, PI), the Theme-based Research Scheme (T44-410/21-N to U. Maurer, PC, C. McBride, PC, and T. Inoue, Co-PI), and the General Research Fund (Project Number 14617721 to T. Inoue, PI) from the Hong Kong Special Administrative Region Research Grants Council