Do you have to assess reliability for non-composite scores? I feel like this is probably a really stupid question but I'm having issues finding the answer. When I think of reliability, normally I think of several items on a test that are compiled into some kind of sum score, with which the individual items are tested with a reliability coefficient like alpha, gamma, or omega.
However, if one simply has a singular measure of an item (either a sum, mean, or categorical outcome), there would be nothing to estimate as far as reliability right? For example, the alpha formula is:
$\alpha = \frac{k}{k-1}(1-\frac{\Sigma V_i}{V_t})$
where $k$ is the number of items, $V_i$ is the variance by item and $V_t$ is the total variance of all items. Given this is the case, it wouldn't make sense to run alpha on only one item right? The reason I ask is because measures like weight, inches, etc. are direct and usually don't require assessments of reliability but a single question on a psychological test ("Do you find this situation scary?") may not be so valid but would perhaps be reliable if all other factors are controlled for.
 A: After digging around for an answer, I found a pair of really interesting articles on this subject. The first is a meta-analysis conducted on a single-item job satisfaction scale, which showed that despite having one item, the aggregated correlation was fairly strong (r = 0.63). Another article used a latent-variable modeling longitudinal approach to finding the reliability of a single-item life satisfaction scale, which showed that it was also within reliability thresholds (above the typical .70 alpha cutoff). In either case, the authors note that reliability cannot be measured on these items on their own, but can be aggregated in the ways discussed in the articles to show that they are still dependable across contexts.
Some interesting points made by these articles that I never considered are the need for such items in instances where respondent burnout can be high. They also mention in at least one of these articles that the choice of a multiple-item and single-item scale is dependent on the factors that necessitate their use. I have cited both studies below.
Citations:

*

*Wanous et.al., 1997


*Lucas & Donnellan, 2012
