An Experimental Comparison of Clustered and Scattered Scale Items

For at least the last 30 years social scientists have argued whether items in a scale should be clustered together or scattered apart. Psychologists designing personality inventories and social psychological scales have usually recommended the randomization of items. The Minnesota Multiphasic Personality Inventory for example contains 550 items made up of 9 psychiatric discriminant keys and four control or validity keys. Items are randomized throughout the test. The main purposes of randomization on these types of scales are to (1) help disguise the purpose of the scales and thus reduce intentional faking; (2) spread each scale across the test so transitory conditions (e.g., moods) and test reactions (e.g., fatigue) will fall equally on all scales; and (3) eliminate the possibility of order effects between scales by randomizing the constituent items.1 Political scientists and sociologists on the other hand almost always cluster items together on their much shorter scales. This is usually done because of the face comprehensibility of presenting similarly formated and related items together and because complete randomization of items covering differing topics and using various formats does not seem practical.2 Despite these clear differences in ordering procedures between disciplines, general texts on survey methodology and test/