magnitude estimation Technique
arnold on Sep 16, 2008
Schriesheim and Castro (1996) did the study on this.
THE BASIC ARGUMENT:
Prior literature mainly used ‘sometimes’ as the referent. So what would happen if the referent that the frequency expressions were tied to was changed? Would it enhance confidence placed on the magnitude estimation technique (MET) through proof of invariance or undermine the confidence placed on MET through indication of existence of variance between different results to different referents? The study was carried out to test for invariance. A result indicating invariance would enhance the confidence placed on MET as ‘a procedure for scaling equal interval survey response alternatives’. The hypotheses was tested using two referents (‘sometimes’ and ‘fairly many times’).
Results: When utilized as the referent, the expression ‘sometimes’ resulted in higher statistically significant results. This result was not the same when it was not the referent.
The ppposite was true for ‘fairly many times’. Even though there was a statistically significant difference between the subgroups , this was not the case when it was the referent. The subsample that utilized ‘sometimes’ as referent produced only two statistically significant differences (not very often and not often). On the other hand, ‘fairly many times’ obtained 13 statistically significant differences (all expressions but very often, not at all, and never)
1) Single referent use in MET when constructing scale expression for survey responses should be avoided. Multiple referents would be a better idea.
2) The scale values used in single referent MET studies should not be relied upon in the selection of alternatives for equal response anchors.
All the prior literature on the issue used ‘sometimes’ as the referent. This study through use of two referents (‘sometimes’ and ‘fairly many times’), concludes that different standards yield different results, which indicates lack of invariance.
Schriesheim, C.A. & Castro, S.L. 1996. Referent Effects on the magnitude estimation scaling of frequency expressing for response anchor sets: An Empirical Investigation. Educational and Psychological Measurement; 56: 557-569
Can anyone make this more clear for me?