Large-Scale Assessments using EdSurvey R Package

AERA 2024 Workshop by AIR

Clustering methods

1. jackknife repeated replication (By default TIMSS and NAEP)
2. Taylor series approximations
3. Hierarchical linear models / Mixture model

Sampling (takeaway)

1. Schools are sampled with probability proportional to size: larger schools are more likely to be selected
2. Thus, sample weights should be used to correct the sampling bias
3. Student weight is the inverse of the probability of selection

W_{final} = \frac{1}{P_{school}*P_{student}*P_{adj}}

Where

1. P_{student} is the probability of one student being selected within one school
2. P_{school} is the probability of one school being selected
3. P_{adj} are non-participation adjustments

Plausible Values

1. the distribution of latent scores for each individual is estimated by both IRT and latent regression with survey variables¹

¹ For each student, there are hundreds of context factors, so PCA was used for dimension reduction

EdSurvey-GPT

EdSurvey-GPT is a chatbot.

Takeaways:

1. Examples of using functions of EdSurvey
2. Code debuggging
3. Data download using R functions