concept of values. Of course it also is possible that response set is operating across the pro-worded items; if this were the case, reliability as determined by coefficient alpha could be high, even though many responses were invalid (in this instance, con sentiment misrepresented as pro sentiment). However, the format of the response options-to what extent does respondent feel one way or the other, pro or con-makes response set much less likely than other formats (e.g., agree-disagree). For assessing the potential usefulness of the indicator we do just rely on reliability (there is no infallible measure, since even test-retest reliability can be contaminated by true change), but look, additionally, at how the scale performs in relation to other variables. 4The urban Costa Rica sample was drawn from sample frame prepared by Professor G6mez, already used and refined in six national probability samples as well as several smaller studies. In the present investigation, in order to retain comparability with the urban sample of New York, only two of the four strata in the sample frame were used, the metropolitan area of San Jose and the urban areas of provincial capitals of Alajuela, Cartago, and Heredia. Eliminated were the two rural strata. Each stratum was in turn stratified according to an index of socioeconomic status, based upon housing characteristics and ownership of electrical artifacts as reported in the 1973 national census. Primary sampling units This content downloaded from 157.55.39.221 on Sun, 06 Nov 2016 05:48:04 UTC All use subject to http://about.jstor.org/terms ANTISYSTEM POLITICAL BEHAVIOR 251 can survey, conducted in Guadalajara during July and August 1979. This purposive sample, with total N of 169, was drawn from six middleclass sections of Guadalajara. Planned by Seligson and John A. Booth, the study was conducted with the institutional support of the University of Arizona Guadalajara Summer Program. In the Guadalajara survey ten-point response format was used in place of the seven-point format, and, because of space limitation, only items C through G were administered. The reliability analysis for the Urban Costa Rica and Guadalajara samples is also given in Table 2. The Political Support-Alienation scale shows satisfactory reliability in the Costa Rican sample and achieves an even higher level of reliability in the Mexican sample. These surveys also included the five Trust in Government items, administered according to the same format used in the National Election Studies carried out by the Survey Research Center of the University of Michigan.5 For reliability analysis, we dichotomized them following the procedure of Miller (1974, p. 953), with the most negative response (see Table 1) scored as 1 and the other categories scored O. In the New York City general public survey, only subsample consisting of 372 respondents received the Trust in Government items. Among the 284 general public respondents could be scored on all items, Table 3 shows that the reliability coefficient is .70, while in the university sample it is even lower, reaching level of only.66. These are low, though unacceptable (according to customary practice in basic research), levels of reliability. Interestingly, respondents from academia appeared to have more trouble with these items than did the general public. From Table 3 one also sees that the Trust in Government items have similarly low, reliability in two of the Latin American samples. If we designate values less than .6 as unacceptably low reliability, the Costa Rican sample approaches this level. The Guadalajara sample shows somewhat higher reliability-but still in the .6-.7 range. Also, the blue-collar Mexican border-cities analyzed by Seligson (forthcoming) showed reliability coefficient in the .6-.7 range (.63). were then selected at random using the PPS method. Within each unit all dwellings were visited, and list of all residents eighteen years and over was prepared. The final selection of respondents was then drawn at random from these lists. I Since 1974 the SRC has made minor change in the wording of the last two items in the series. The smart item was revised by deleting who usually know what they are doing from the first part of the item, while the crooked item was revised by inserting the words a little before the first use of crooked and deleting crooked after not very many are. We have used the original format to retain compatability with the earlier SRC work as well as with some of our own studies. These changes do appear to have increased the reliability of the SRC data. This content downloaded from 157.55.39.221 on Sun, 06 Nov 2016 05:48:04 UTC All use subject to http://about.jstor.org/terms
Read full abstract