January 15, 2020
Comparing PAM 10 and PAM 13
The measurement power of PAM versions is assessed in four key manners. (Download full comparison)
Difficulty structure: These item calibrations show how difficult it is for respondents to agree to an item.
Person reliability: Person reliability measures how effectively a series of questions separates subjects on the basis of the measure. Person reliability using Rasch analysis is equivalent to Test Reliability (Spearman) or sample reliability. The low value is called “real” reliability and assumes that any variation of a person’s measured activation from what the model expected is due to measurement error and not random error. This is never fully the case. The high value is called “model” reliability and assumes that none of the variability between a person’s expected activation and their measured activation is due to measurement error; it is all just random error. Again, this is never true. We know that the actual reliability is somewhere between these two reliability values. High person reliability indicates that there is a high probability that a person who is highly activated will receive a high activation score using the PAM, and conversely, a lower score will indicate lower activation. Person reliability near .80 or better is desired.
Cronbach’s Alpha: Internal consistency reliability. People think it tells us the proportion of total variance in person scores that is true score rather than measurement error. But as explained about, it does not. Industry standards are at least .7 and at least .9 if you are tracking people over time with the measure.
Predictive power: How well does the PAM predicting behaviors and outcomes.