An Effective Approach to the Repeated Cross‐Sectional Design

Abstract

Repeated cross‐sectional (RCS) designs are distinguishable from true panels and pooled cross‐sectional time series (PCSTS) since cross‐sectional units (e.g., individual survey respondents) appear but once in the data. This poses two serious challenges. First, as with PCSTS, autocorrelation threatens inferences. However, common solutions like differencing and using a lagged dependent variable are not possible with RCS since lags for i cannot be used. Second, although RCS designs contain information that allows both aggregate‐ and individual‐level analyses, available methods—from pooled ordinary least squares to PCSTS to time series—force researchers to choose one level of analysis. The PCSTS tool kit does not provide an appropriate solution, and we offer one here: double filtering with ARFIMA methods to account for autocorrelation in longer RCS followed by the use of multilevel modeling to estimate both aggregate‐ and individual‐level parameters simultaneously. We use Monte Carlo experiments and three applied examples to explore the advantages of our framework.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 106 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.