Abstract
The article reports the results of a Mokken Scale Procedure (MSP) developing a hierarchical cross-national scale gauging attitudes toward illegal immigration, and a subsequent qualitative cross-national assessment of this scale. Responses to a 20-item Likert-type-scale were collected in two national representative surveys in the Netherlands and New Zealand. The MSP analysis yielded a cumulative scale with the same four items for each with an acceptable ‘scalability’ in both samples, of H > .40. This cross-national four-item scale was evaluated by means of in-depth interviewing nine participants in the Netherlands and 15 participants in New Zealand. Analysis of the interviews shows that individually ranking the items of the scale is similar to the rank order generated by MSP, but the individual evaluation of the degree of negativity of items strongly depends on the way illegal immigrants are framed by subjects. The contribution of a mixed methods approach is discussed. It is suggested that the present quantitative measuring instruments measure a general average attitude, which may be likened to a measure of the average temperature of a country at a certain moment, while qualitative evaluations vary with the way illegal immigrants are framed by individuals, like variations in local temperature
Notes
1. The rule of non-intersection (meaning that the IRFs should not intersect) is tested by looking at Rest score, Rest split and the P-matrix, and by erasing the weakest link in every step. Items where the indicator exceeds the level of 80 can be seen as worrisome, leading to a concern about the model assumption. Items with levels between 40 and 80 are questionable, and values lower than 40 can arguably be ascribed to sampling errors (Molenaar & Sijtsma, Citation2000, pp. 12–49, 74). Only the items with the highest critical value are deleted, one at a time. After deleting one item the critical non-intersection levels of all the other items will improve.In order to test homogeneity, a H value of < 0.3 is considered as unscalable, between 0.3 and 0.4 as representing a weak scale, between 0.4 and 0.5 a medium scalability and strong scalability when ranging from 0.5 to 1.0 (Molenaar & Sijtsma, Citation2000, p. 12). To decide about monotonicity (M): a summarizing value crit per item needs to be lower than 80, and preferably, lower than 40 (Molenaar & Sijtsma, Citation2000, p. 74). The scale reliability (RHO) => 70 (Molenaar & Sijtsma, Citation2000, p. 15).