Publication Type: Conference Paper/Unpublished Manuscript
Review Method: Peer Reviewed
Abstract: In the Respondent-Generated Interval (RGI) technique, respondents are asked to recall a numerical quantity by giving both a best estimate and an interval that bounds the values that the respondent thinks the recalled quantity might take. The RGI technique then estimates a population mean using a weighted average of the values given by the respondents. The weights are functions of the lengths of the intervals supplied by the respondents; longer intervals, assumed to represent less accurate recall, generate smaller weights than do shorter intervals, assumed to represent more accurate recall.

In this paper we examine this assumed relationship between accuracy of recall and length of interval. We use data from record-check surveys carried out on two college campuses. If the assumed relationship does not hold overall, we examine patterns in the data to attempt to identify subsets of respondents for which the relationship does hold.

Publication Type: Paper/Poster Proposal
Abstract: Measures of election poll accuracy, starting with Mosteller’s pioneering work in 1949, have always involved “point estimates” of error, using a single set of predicted vote percentages. A known drawback of any such measure is that it must ignore "undecded" voters or make an arbitrary assumption about how to allocate them. A new measure is proposed, which uses an “interval estimate” to account for every possible allocation of undecided voters, rather than choosing to ignore undecided voters or allocate them arbitrarily. Advantages and disadvantages of this measure are discussed.

Publication Type: Conference Paper/Unpublished Manuscript
Abstract: Criminologists widely acknowledge an age-crime curve: Criminal offending onset peaks between ages 8-14, prevalence peaks between ages 15-19, and desistance peaks between ages 20-29 (for a review, see Farrington, 2003). There is less agreement, however, on the explanation for these patterns. This paper elaborates Akers and Lee’s (1999) supposition that variables specified in social learning theory can account for the relationship between age and criminal or deviant behavior, testing derived hypotheses on data from the Racine birth cohort. Giving attention to each of Farrington’s 24 key empirical and theoretical developmental and life-course issues, this paper presents the argument that social learning and operant conditioning concepts form the underlying explanation for age differences in crime and deviance. This paper posits that onset, continuance, and desistance of criminal or deviant behavior, manifest in an age-crime curve, are functions of differential association, variable-interval reinforcement, and extinction.

Publication Type: Conference Paper/Unpublished Manuscript
Review Method: Peer Reviewed
Abstract: Some New Directions for the Respondent-Generated Intervals Protocol

LiPing Chu, S. James Press, and Judith M. Tanur

The Respondent-Generated Intervals protocol (RGI) has been used to have respondents answer a factual question by giving not only a point estimate but also bounds within which they feel it is almost certain that the true value of the quantity being reported upon falls. Some new thinking aims to elaborate the RGI protocol with the goal of improving the accuracy of the estimators derived from the protocol.

Because point estimates of respondents who give short intervals are weighted more heavily in the Bayesian RGI estimator than are point estimates of respondents who give longer intervals, it is advantageous to encourage respondents who are more accurate to give shorter intervals and respondents who are less accurate to give longer ones. Mechanisms to direct confident (and thus presumably accurate) respondents to give shorter intervals and less confident (and thus presumably less accurate) respondents to give longer ones have been suggested, and such mechanisms tested in an experimental design that varies the instructions about how respondents should construct their intervals. This paper reports on this experiment, giving preliminary results.

Publication Type: Conference Paper/Unpublished Manuscript
Review Method: Peer Reviewed
Abstract: Respondent-generated intervals (“RGI”), a question item protocol promoted by statisticians Press and Tanur, promises to have a positive impact on survey data quality, both by reducing item nonresponse and by improving the accuracy of estimation.

In this experiment, we put this method (RGI) to test with a sample of electronics engineers each reporting on a unique integrated circuit (chip) design developed in 2002.

The treatment version of the questionnaire (a self-administered web survey) encourages respondents to provide a self-generated interval if they are unsure of the number for the physical quantity requested (e.g., number of transistors on the chip, number of bytes of embedded memory, etc.). Respondents in the control group receive no such encouragement and are just asked to provide the relevant number.

1) This experiment seeks to answer a question: Does the suggestion to provide an RGI reduce item nonresponse?

2) The second objective of this experiment is to contribute to the discussion of whether RGIs can provide more accurate estimates than are obtained by merely asking respondents for one number. Research has shown that when asked for a number whose exact value is unknown, people resort to an “anchoring-and-adjustment heuristic” which often results in biased values. The bias is due to the original anchor value, which acts as a tether preventing sufficient adjustment. Because RGIs promote the reporting of at least two values, this very process might well break the “pull” which the original anchor value has on a respondent’s ability to provide an unbiased estimate.

The purpose of this research is to elicit questions dealing with ways to help respondents reporting factual data to provide unbiased estimates of target quantities. The biases we are talking about here are cognitive in nature and not emotional (e.g., social desirability).