Validity and Transparency in Quantifying Open-Ended Data

Abstract

Quantitatively coding open-ended data (e.g., from videos, interviews) can be a rich source of information in psychological research, but reporting practices vary substantially. We provide strategies for improving validity and reliability of coding open-ended data, and investigate questionable research practices in this area. First, we systematically examined articles in four top psychology journals (N = 956) and found that 21% included open-ended data coded by humans. However, only about one-third of those articles reported sufficient details to replicate or evaluate the validity of the coding process. Next, we propose multi-phase guidelines for transparently reporting on the quantitative coding of open-ended data, informed by concerns with replicability, content validity, and statistical validity. The first phase involves research design, including selecting data and identifying units reliably. The second phase includes developing a coding manual and training coders. The final phase outlines how to establish reliability. As part of this phase, we used data simulations to examine a common statistic for testing reliability on open-ended data, Cohen’s kappa (κ), and found that it can become inflated when researchers repeatedly test inter-rater reliability or manipulate categories such as by including a missing data category. Finally, to facilitate transparent and valid coding of open-ended data, we provide a pre-registration template that reflects these guidelines. All of the guidelines and resources provided in this paper can be adapted for different types of studies, depending on context.

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 — 16 citations exist upstream.

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