Reliability (Inter-Item-Correlation) Analysis
Cronbach's alpha is a statistic frequently employed to indicate the reliability (or internal consistency) of a number of items that supposedly form a scale. Stata's procedure offers an interesting possibility to compute the scale immediately from your data.
alpha trust1 trust2 trust3 trust4 trust5, item generate(trust)
will compute alpha and, in addition, will compute the scale and save the values in the variable whose name is given in parentheses after option
generate (abbreviated as
g). Note that Stata in computing the scale will deal with missing values in an interesting way: The scale will be computed as a standardized variable; and if a case has one or several items missing, the standardized variable will be computed using only those items that have valid values. Note that "standardization" here simply means adding up the scores and dividing the sum through the number of items that entered into the computation. – The
i) option requests correlations between each item and the rest of the other items, plus the alpha value that would be obtained if an item were removed.
There is a further option,
c) which will delete cases with missing values from the analysis. In this case, scores will be computed only for complete cases.
By default Stata will assume that a variable that correlates negatively with the other variables is a "reversed" item; it will re-reverse the values of such a variable for the analysis, as it were. To suppress this, add the option
a. You may also specify the variables to be reversed (and, by implication, those not to be reversed) with the option
r, such as in
Note that Stata will also "reverse" the item in computing the scale as described above. But, to put it bluntly: the way it works is bullshit. What Stata does to reverse an item is to change its sign. That is, if you have, say, a Likert-scaled item ranging from 1 to 5, a value of 5 will enter as -5 in the computations. What you would want, however, is that 5 will be changed to 1, and vice versa.
© W. Ludwig-Mayerhofer, Stata Guide | Last update: 30 Nov 2010