Premium
Making Theoretically Informed Choices in Specifying Panel‐Data Models
Author(s) -
Ketokivi Mikko,
Bromiley Philip,
Awaysheh Amrou
Publication year - 2021
Publication title -
production and operations management
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 3.279
H-Index - 110
eISSN - 1937-5956
pISSN - 1059-1478
DOI - 10.1111/poms.13347
Subject(s) - underpinning , econometrics , statistical inference , specification , inference , hausman test , panel data , computer science , test (biology) , statistical model , statistical hypothesis testing , random effects model , fixed effects model , economics , statistics , mathematics , medicine , paleontology , civil engineering , meta analysis , artificial intelligence , machine learning , engineering , biology
We argue that in analyzing panel‐data econometric models, researchers rely excessively on statistical criteria to determine model specification, treating it primarily as a matter of statistical inference. This inferential emphasis is most obvious in the common practice of using statistical tests (e.g., the Hausman test) to choose between fixed‐ and random‐effects specifications, often ignoring the assumptions underpinning these tests. For instance, the Hausman test depends on the true within‐panel (longitudinal) and between‐panel (cross‐sectional) parameters being equal. This assumption is often not justified, because longitudinal and cross‐sectional variances and covariances may manifest different underpinning mechanisms. In addition to different mechanisms often resulting in different variables determining within and between effects, within and between variables may also have different meanings. To help researchers make theoretically informed choices, we formulate five questions that can guide researchers to think of model specification in a theoretically rigorous way. We examine these issues with examples from both general management and operations management research. Importantly, we argue that addressing the questions regarding model specification must involve primarily theoretical and contextual judgment, not statistical tests.