← Data Science for Educators

Digital Assessment Mode Effects

Did switching NAEP from paper to computer tests in 2017 distort the scores, and does this confound pre-post COVID comparisons?

Causal (instrumentation threat to validity)

What is this question really asking?

In 2017, NAEP moved from paper booklets to tablets. If students who rarely used computers at home suddenly took a high-stakes test on a screen, their scores might have dropped — not because they knew less, but because the format was unfamiliar. This is called a mode effect. The concern: if the format change artificially lowered scores right before COVID, then some portion of the "COVID learning loss" might actually be a measurement artifact — the testing instrument changed, not the students.

Why classification matters. This is an instrumentation threat to internal validity — a concern that the measuring tool itself changed between measurements, making before-and-after comparisons unreliable.

How you would investigate it

  1. Compare 2015 (paper) to 2017 (digital). Look at the score change for all students at the transition. If there was a sharp, one-time drop in 2017 that did not continue in 2019, that pattern suggests a mode effect rather than a real decline.

  2. Split by technology access. Group students by whether they had a computer at home, internet access, and experience with tablets at school. If students without home devices showed larger drops in 2017, the effect is differential — it hits some groups harder than others — which is strong evidence of a mode effect.

  3. Estimate the magnitude. If the mode effect was roughly 2 points, then the "true" COVID-era decline was 7.9 minus 2 = 5.9 points. Still a serious drop, but not historically unprecedented. This is a sensitivity analysis: you recalculate your main finding under different plausible assumptions to see if it holds up.

  4. Check if the effect was temporary. By 2019, students had more device experience. If the 2017 drop partially reversed by 2019, the mode effect was transient and does not affect COVID-era comparisons much.

Data you would need

What Source Example
Scores for 2015, 2017, 2019 NAEP Data API 2015: paper; 2017: first digital; 2019: second digital
Student device access NAEP student survey (B034405–B034407) Desktop/laptop, tablet, smartphone at home
Teacher technology training NAEP teacher survey (T131851) "Have you had training on classroom tech?"
School device availability NAEP school survey (TECHAVS) Device types available for student use
NCES mode-effect study 2017 technical documentation Official analysis of DBA transition

NAEP API variables: B034405–B034407, B034401, B034801, T131851, T132101–T132103, TECHAVS

Analytic method: Interrupted time series design with 2017 as the interruption point. Device access as an interaction term. Sensitivity analysis recalculates COVID-era declines excluding the estimated mode effect. Note: the API sample field (R2 = accommodations not permitted, R3 = accommodations permitted) tracks accommodation policy, not assessment mode. Mode effects must be inferred from score trajectories rather than a direct mode flag.