|
|
Further Information
Primer of Statistical Tools
|
A scale combines responses to a series of items into one single value. Scales can be the sum of dichotomous choices, value rankings, or other choice responses. What's important is that the items work together to build a coherent scale. The first step is to write items that theoretically belong with the concept for which the scale is being designed. Following theoretical construction the coherence of items must be statistically checked. One way to evaluate this coherence is to look at the correlation between each item used in a scale and the scale without that item. If the correlation is positive and significant then the item can be judged to belong in the scale. In the following example five items were proposed as working together. The items A, B, and C each have a reasonable correlation (.29 or .28) to a scale that summed up the other four items. Item D has a lower correlation, however, since it is positive it is not as great an issue as item E. Item E has a very strong negative correlation to the scale. It can be judged that this item does not fit in the scale.
Two responses are possible based on the -.56 correlation. The item can be removed from the scale and the correlation checked. Without E the correlations for items A to D will increase substantially. At this point item D can be evaluated and a further judgment made. In some cases tool developers include null or non-related items in order to minimize respondent manipulation and/or increase independence among scale. It is interesting to note that the strong negative correlation of item E may indicate that it is connected to the idea, but that connection is reversed or negative. Some scales are built with both positive and negative items. In these cases the coding of some items is reversed when they are included in a scale. |
|
