Alternative of Cronbach alpha (for single item measures)

[Sonnen Blume]

Hi,
I've data on patients subjective health measured by a single question "How is your health today". I've to check the validity of this question so I tried 'Cronbach alpha', but it doesn't work for single item measures. Is there any alternative of it in Stata 14 that can do the trick?
Thanks.

[Clyde Schechter]

No, not in Stata, not in anything. Cronbach's alpha measures the extent to which the responses to items in a scale tend to agree with each other. You can sort of think of it as a mass correlation coefficient of the responses with each other. When there is only one item, there is no possibility of agreeing or disagreeing. Internal consistency is only definable with 2 or more items.

Also, Cronbach's alpha is not a measure of validity. Validity means that the item is a measure of the construct you want to measure: it means that the measure would agree with a reference measure of the same construct. Cronbach's alpha just measures the extent to which the various items in the scale agree with each other, but says nothing about how they would compare to a reference measure of any construct. If a scale has a good Cronbach's alpha then it is a consistent measure of something, but it may or may not be a valid measure of the particular construct you were hoping to measure.

If you had repeated observations of this question on the same person, you could do an intra-class correlation to determine its consistency under repeated measures (aka test-retest reliability). That would be somewhat analogous to a Cronbach alpha. But to establish validity, you would have to show that the response to this question agrees with some reference measure of health. If you search the literature, I believe you will find that others have, in fact, shown that self-reported general health, as assessed in this way, is, in fact, positively associated with residual life expectancy and certain other accepted measures of health.

[Sonnen Blume]

That's a very clear interpretation as always. Thanks a lot Clyde!

In this case its not test-retest reliability, but its an important thing to know.