Researchers retain theoretically justified variables even when statistically insignificant.

Evidence

In order to validate the measurement model, the first test is for multicollinearity between the items of the LV. The PLS model was calculated with the SmartPLS 2.0 (M3) software package (Ringle et al., 2007). After looking at the variance inflation factor (VIF) values for all items including the items for the exogenous LV, multicollinearity can be excluded (all VIF < 10). Secondly, indicator relevance is tested by first looking at the values of path weights for each indicator and secondly running a t-test for significance. Figure 5.4 shows the results of the structural model of path dependence through CPA and states the path weights and R2 value. Only the path coefficient for the factor ‘Degree of Political Activity’ is not significant. For theoretic reasons, we decided to leave the factor included in the model (Lohmöller, 1989). Also, we find that all path coefficients have the predicted sign. Hence, Hypothesis 1 can be approved. — Georg Schreyögg and Jörg Sydow, The Hidden Dynamics of Path Dependence, 409

Categorization

• DEFINE: Demonstrates a methodological decision rule: significance testing is bounded by theoretical rationale, not the reverse.
• XREF: Connects to broader philosophy-of-science debates about whether theory or data should adjudicate variable retention in structural models.