Using Corpus Analyses to Help Address the DIF Interpretation: Gender Differences in Standardized Writing Assessment
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Addressing differential item functioning (DIF) provides validity evidence to support the interpretation of test scores across groups. Conventional DIF methods flag DIF items statistically, but often fail to consolidate a substantive interpretation. The lack of interpretability of DIF results is particularly pronounced in writing assessment where the matching of test takers' proficiency levels often relies on external variables and the reported DIF effect is frequently small in magnitude. Using responses to a prompt that showed small gender DIF favoring female test takers, we demonstrate a corpus-based approach that helps address DIF interpretation. To provide linguistic insights into the possible sources of the small DIF effect, this study compared a gender-balanced corpus of 826 writing samples matched by test takers' performance on the reading and listening components of the test. Four groups of linguistic features that correspond to the rating dimensions, and thus partially represent the writing construct were analyzed. They include (1) sentiment and social cognition, (2) cohesion, (3) syntactic features, and (4) lexical features. After initial screening, 123 linguistic features, all of which were correlated with the writing scores, were retained for gender comparison. Among these selected features, female test takers' writing samples scored higher on six of them with small effect sizes in the categories of cohesion and syntactic features. Three of the six features were positively correlated with higher writing scores, while the other three were negative. These results are largely consistent with previous findings of gender differences in language use. Additionally, the small differences in the language features of the writing samples (in terms of the small number of features that differ between genders and the small effect size of the observed differences) are consistent with the previous DIF results, both suggesting that the effect of gender differences on the writing scores is likely to be very small. In sum, the corpus-based findings provide linguistic insights into the gender-related language differences and their potential consequences in a testing context. These findings are meaningful for furthering our understanding of the small gender DIF effect identified through statistical analysis, which lends support to the validity of writing scores.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it