Suppose a scale has two items that seem very similar. For instance, suppose you have a survey with two items that people rate their level of agreement with (e.g., 1 = strongly disagree, 7 = strongly agree), but the two items look almost identical and do not appear to be meaningfully different.
What problems, if any, may arise from using items that may be redundant or "too similar"? And are there any heuristics or guidelines for making judgment calls about item similarity?
Outside whatever quantitative means one might use to assess item redundancy, what should one do if the items don't meet these criteria, but still seem like they aren't meaningfully different? I would worry that any reliable difference in participant response isn't an indication that the two items are assessing non-overlapping contours of a construct, but that there's just something idiosyncratic about the difference in response that has nothing to do with capturing the construct in question.
I would absolutely be interested in any articles on this topic as well. Thanks!
Hi Lance,
From an IRT standpoint, items of an ideal measurement should capture the whole continuum of a construct. There should be items sensitive to people who are low/mid/high in the construct, so that people varied in the construct can be differentiated.
Realistically, no scales and inventories are that ideal and they usually capture only a section of the continuum, such as in the "high" ground. In other words, items are concentrated to reliably differentiate people who are high in the construct, but not for low or mid. Consequently, having a redundant item may suggest that the scale has relatively more weight in a particular section of the continuum. You can conduct an IRT to have a look.
A scale with redundant items usually has a high alpha. But too high may not be a good sign since a scale should capture some more variations. You can take a look at the correlational matrix amongst items when you examine the alphas.
Sometimes the similarity in wording and the order of items may lead to some quantitative indications for redundancy even there are none.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling