Adjusting Likert Item Scoring After Finding Very Low Reliability
A Story by Larry Nelson, Curtin University

18 August 2010


The data are from the administration of a 48-item attitude scale.  All items used a 1 – 5 Likert scale, with 5 = “strongly agree”.


The original control “cards” used by an overhasty graduate student:


The reliability was terrible (especially as this was a well-known affective scale)!


Several items had negative correlations:


The bottom of the Stats1f report was revealing:


The IStats results were even more revealing (note negative p-comp1 loadings) :


Larry’s new CCs lines in green (note the *pol line):


Vastly-improved reliability noted:


No negative correlations now (note the pol. column):


The bottom of Stats2f was better than Stats1f:


The new IStats report was much better (note movie shows how to get IStats):


What the overhasty graduate student had overlooked:

Many of the items in the scale were negatively-worded. For example, the sample item shown on line 12 above (item “14”) should be reverse-scored. The graduate student had simply failed to use a *pol line to correct the scoring for those scale items which were negatively-worded.




A scree plot is easy to get (note movie shows how):


There are a few System settings which are relevant:


Se Tutto!