Calling Models With Causal Indicators “Measurement Models” Implies More Than They Can Deliver

Click to increase image sizeClick to decrease image size Notes1 This definition raises the intriguing possibility of other types of links. For example, we might conceive of "enjoyment of reading" as a correlational indicator of extraversion: More extraverted people tend to spend less time reading, but reading is neither a cause nor an effect of extraverted personality.2 The only way that causal indicators can affect the variance of the latent variable or its covariance with other variables in the model is if there is some misspecification in the relations between causal and effect indicators; in particular, if the variance shared among effect indicators (i.e., the latent variable) does not fully mediate the relations between each causal indicator and each effect indicator. Such misspecification can cause the reflective factor loading estimates to shift to account for the misspecification. When the model is correct, however, causal indicators can be removed from the model without affecting the latent variable.