Table 8 - uploaded by Daniel Mcfadden
Content may be subject to copyright.
Source publication
The unfolding bracket method for eliciting quantitative information in economic surveys is effective in reducing item non-response and outliers, but is vulnerable to bias induced by anchoring. To test for anchoring effects, this study introduced an experimental module in the AHEAD (Asset and Health Dynamics of the Oldest Old) panel in which the unf...
Citations
... One strand examines the intersection between questionnaire format and survey biases. For example, Hurd et al. (1998) show that anchoring effects may account for significant differences in responses when using brackets versus open-ended questions on savings and consumption data; Frykblom and Shogren (2000) argue that anchoring effects in discrete-choice questions may result from survey framing rather than from the dichotomous choice per se. Hitczenko (2015) finds strong evidence of sequential anchoring in Likert scale questions, resulting in approximately 13% of responses incorrectly rated for selected questions. 4 We extend this body of literature by examining the role of anchoring bias in retrospective data for objective and subjective measures in a low-income setting. ...
Self-reported retrospective survey data is widely used in empirical work but may be subject to cognitive biases, even over relatively short recall periods. This paper examines the role of anchoring bias in self-reports of objective and subjective outcomes under recall. We use a unique panel-survey dataset of smallholder farmers from four countries in Central America collected over a period of three years. We exploit differences between recalled and concurrent responses to quantify the degree of mental anchoring in survey recall data. We assess whether respondents use their reported value for the most recent period as a cognitive heuristic when recalling the value from a previous period, while controlling for the value they reported earlier. The results show strong evidence of sizeable anchoring bias in self-reported retrospective indicators for both objective measures (income, wages, and working hours) and subjective measures (reports of happiness, health, stress, and well-being). We also generally observe a larger bias in response to negative changes for objective indicators and a larger bias in response to positive changes for subjective indicators.
... se the most when a toll is introduced. For Eliasson and Mattson (2006), the impact of a toll on this population category would increase as their incomes and capacities to adapt decrease. Their capacities to adapt depend on their travel purposes. The work purpose allows them the least latitude. Furthermore, agents can be strategic in their response. Hurd et al. (1998) point out the strategic misrepresentation of the respondent when he anticipates a possible link between his response and some economic outcome in which he has an interest. In our survey, agents will be strategic because we present them with a road pricing scheme they will have to pay and which is quite close to their situation. We don't ...
Various methods of restricting automobile traffic, by price (tolls) or by quantity (odd/even license plates or limited days of traffic), are tested in a survey (N ∼ 400) about attitudes toward traffic restrictions in Lyon, France. Ordered probit models with random-effects panel allow us to estimate the survey respondents’ perceptions of these methods, as well as the roles of individual socio-demographic characteristics in the formation of these perceptions. Both the restriction of automobile traffic and its regulation by congestion (waiting in line) are widely considered unjust by the respondents, regardless of whether they work and whether they are drivers or non-drivers. Their attitudes towards tolls justified by the pollution caused by automobile traffic are less negative. As regards compensation, in addition to emergency vehicles and those that transport people with limited mobility, respondents believe car-pooling ought to benefit of a toll exemption. The support for a reduced rate for low-income users shows a concern for justice to which it will be necessary to respond. The respondents’ socio-professional status, level of education, car use or non-use, and residence inside or outside of the toll zone clearly play a role in their perceptions of these methods of regulation and compensation.
... Thus, weather reports and pilot communications include a form of implicit interval , e.g. when meteorologists report that the wind blows from the " north means that the wind direction lies in the interval [337.5., 22.5.]. " This kind of rounding is also prevalent in many types of surveys; it is well-documented that individuals often round their answers to open-ended survey questions, see Rosch (1975), Schaeffer and Bradburn (1989), Huttenlocher, Hedges and Bradburn (2008), Hurd et al. (1998), Hobbs (2004 rounds, add an interval option to the current point-type question about a certain probability. The basic problem is that when a respondent is to report a point, he sometimes round it to describe a sentiment that really is an interval. ...
We analyze an approach to quantitative information elicitation in surveys that includes many currently popular variants as special cases. Rather than asking the individual to state a point estimate or select between given brackets, the individual can self-select any interval of choice. We propose a new estimator for such interval censored data. It can be viewed as an extension of Turnbull's estimator (Turnbull (1976)) for interval censored data. A detailed empirical example is provided, using a survey on the valuation of a public good. We estimate survival functions based on a Weibull and a mixed Weibull/exponential distribution and prove that a consistent maximum likelihood estimator exists and that its accuracy can be consistently estimated by re-sampling methods in these twofamilies of distributions.
... Several models for what the answers to bracket questions look like if they suffer from anchoring have been introduced and estimated, using different sources of experimental data with random entry points. Examples are Hurd et al. (1998), Herriges and Shogren (1996), and Cameron and Quiggin (1994). A comparison between competing models for anchoring, however, has not yet been performed. ...
... being asked an open question (see Hurd et al., 1998). Although in the first sub-sample initial non-response is substantial and respondents to the open-ended question are probably not a random sub-sample, it will be shown that comparing the combined distribution of open responses and bracket responses in the first sub-population with the distribution in the sub-population of immediate bracket responses provides a test for certain types of anchoring. ...
... For estimates of objective quantities as well as willingness to pay for public goods, they find that, for high entry point values, the proportion of bracket respondents who report that the amount is larger than the entry point value exceeds the fraction of open-ended answers larger than the entry point value. Hurd et al. (1998) specify a parametric model that captures this phenomenon in a symmetric way, biasing probabilities smaller than 0.5 upwards and probabilities larger than 0.5 downward. The idea is that respondents do not compare the true value Y to the entry point E, but instead compare Y to E+ε, where ε is a mean zero error term, assumed to be normal and independent of all other components of the model. ...
In the experimental module of the AHEAD 1995 data, the sample is randomly split into respondents who get an open-ended question on the amount of total family consumption - with follow-up unfolding brackets (of the form: is consumption $X or more?) for those who answer don't know' or refuse' - and respondents who are immediately directed to unfolding brackets. In both cases, the entry point of the unfolding bracket sequence is randomized. These data are used to develop a nonparametric test for whether people make mistakes in answering the first bracket question, allowing for any type of selection into answering the open-ended question or not. Two well-known types of mistakes are considered: anchoring and yea-saying (or acquiescence). While the literature provides ample evidence that the entry point in the first bracket question serves as an anchor for follow-up bracket questions, it is less clear whether the answers to the first bracket question are already affected by anchoring. We reject the joint hypothesis of no anchoring and no yea-saying at the entry point. Once yea-saying is taken into account
... Several models of what the answers to bracket questions look like if they suffer from anchoring have been introduced and estimated, using different sources of experimental data with random entry points. Examples are Hurd et al. (1998), Herriges and Shogren (1996), and Cameron and Quiggin (1994). A comparison between competing models for anchoring, however, has not yet been performed. ...
... In this article we do not analyze reporting errors in open-ended questions, but focus on testing for anchoring and acquiescence errors induced by the bracket questions. We use experimental data on household consumption from the Assets and Health Dynamics (AHEAD) 1995 survey described by Hurd et al. (1998) in which the sample is randomly split between respondents who start with an open-ended question and respondents who immediately go into an unfolding bracket sequence with random entry point. We first develop a test for the null hypothesis that answers to the first bracket question correspond exactly to the respondents' estimates given in open-ended answers, allowing for any form of selective response behavior, that is, for an arbitrary correlation between whether an openended answer or bracket answer is given and the respondent's best estimate of the consumption amount. ...
... For estimates of objective quantities such as willingness to pay for public goods, they find that, for high entry point values, the proportion of bracket respondents who report that the amount is larger than the entry point exceeds the fraction of open-ended answers larger than the entry point. Hurd et al. (1998) specified a parametric model that captures this phenomenon in a symmetric way, biasing probabilities smaller than .5 upward and probabilities larger than .5 downward. In this model, respondents do not compare their initial estimate to the entry point B 1 , but instead compare to B 1 , where E is a mean zero error term, assumed to be normally distributed and independent of other components of the model. ...
We analyze experimental survey data, with a random split into respondents who get an open-ended question on the amount of total family consumption (with follow-up unfolding brackets of the form "Is consumption $X or more?" for those who answer "don't know" or "refuse") and respondents who are immediately directed to unfolding brackets. In both cases, the entry point of the unfolding bracket sequence is randomized. Allowing for any type of selection into answering the open-ended or bracket questions, a nonparametric test is developed for errors in the answers to the first bracket question that are different from the usual reporting errors that will also affect open-ended answers. Two types of errors are considered explicitly: anchoring and yea-saying. Data are collected in the 1995 wave of the Assets and Health Dynamics survey, which is representative of the population in the United States that is 70 years and older. We reject the joint hypothesis of no anchoring and no yea-saying. Once yea-saying is taken into account, we find no evidence of anchoring at the entry point.
... Recent research by Manski (1997, 2004), Gan, Hurd, and McFadden (2003), and others shows that responses to questions of the form "On a scale of 0 to 100, where 0 means no chance, and 100 means certainty, what would you say is the probability of ...?" yield responses that are quite predictive of subsequent behavior. For example, the question asked about mortality risk is generally predictive of subsequent mortality experience McGarry, 1995, 2002;Smith, Taylor, and Sloan, 2001) and more predictive of subsequent savings behavior than life table hazard rates (Hurd, et al., 1998). ...
Measures of households' past behavior, their expectations with respect to future events and contingencies, and their intentions with respect to future behavior are frequently collected using household surveys. These questions are conceptually difficult. Answering them requires elaborate cognitive and social processes, and often respondents report only their “best” guesses and/or estimates, using more or less sophisticated heuristics. A large body of literature in psychology and survey research shows that as a result, responses to such questions may be severely biased. In this paper, (1) we describe some of the problems that are typically encountered, (2) provide some empirical illustrations of these biases, and (3) develop a framework for conceptualizing survey response behavior and for integrating structural models of response behavior into the statistical analysis of the underlying economic behavior. Copyright Springer Science + Business Media, Inc. 2005
... In this paper, several models for anchoring and acquiescence are estimated using experimental data on household consumption from the AHEAD 1995 survey. The specific feature of these data is that the sample is randomly split in respondents that start with an open-ended question -and get follow-up brackets if they do not answer the openended question -and a sub-sample of respondents who are immediately directed to the unfolding bracket questions without being asked an open-ended question (see Hurd et al., 1998). In a companion paper, we have combined the data from the two sub-samples to construct a non-parametric test for the null hypothesis that the first bracket question is always answered correctly (van Soest and Hurd, 2003). ...
... While the authors explain this type of finding with anchoring, the fact that the bias is found for high values only suggests that yea-saying might also play a role. Hurd et al. (1998) specify a parametric model that captures this phenomenon in a symmetric way, biasing probabilities smaller than 0.5 upward and probabilities larger than 0.5 downward. The idea is that respondents do not compare the true value Y to the entry point E, but instead compare Y to E+ε, where ε is a mean zero error term, assumed to be normal and independent of all other variables in the model. ...
... The idea is that respondents do not compare the true value Y to the entry point E, but instead compare Y to E+ε, where ε is a mean zero error term, assumed to be normal and independent of all other variables in the model. Hurd et al. (1998) use the same device also at follow-up bracket questions, with question specific errors that are independent across questions and have a variance that can be different at each question in the unfolding bracket design. Their model is called a gating model, since respondents have to pass a number of gates to reach their final bracket answer. ...
Item non-response in household survey data on economic variables such as income, assets or consumption is a well-known problem. Follow-up unfolding bracket questions have been used as a tool to collect partial information on respondents that do not answer an open-ended question. It is also known, however, that mistakes are made in answering such unfolding bracket questions. In this paper, we develop several limited dependent variable models to analyze two sources of mistakes, anchoring and acquiescence (or yeasaying), focusing on the first bracket question. We use the experimental module of the AHEAD 1995 data, where the sample is randomly split into respondents who get an open-ended question on the amount of total family consumption - with follow-up unfolding brackets (of the form: is consumption $X or more?) for those who answer don t know or refuse - and respondents who are immediately directed to unfolding brackets. In both cases, the entry point of the unfolding bracket sequence is randomized. We compare models in which the probability of a mistake depends on the deviation between the true consumption amount and the entry point amount $X and models in which it does not. We find that allowing for acquiescence bias substantially changes the conclusions on the selective nature of non-response to the open-ended question and on the distribution of consumption expenditures in the population. Once acquiescence bias is taken into account, anchoring in the first bracket question plays only a minor role.
... Several models for what the answers to bracket questions look like if they suffer from anchoring have been introduced and estimated, using different sources of experimental data with random entry points. Examples are Hurd et al. (1998), Herriges and Shogren (1996), and Cameron and Quiggin (1994). A comparison between competing models for anchoring, however, has not yet been performed. ...
... being asked an open question (see Hurd et al., 1998). Although in the first sub-sample initial non-response is substantial and respondents to the open-ended question are probably not a random sub-sample, it will be shown that comparing the combined distribution of open responses and bracket responses in the first sub-population with the distribution in the sub-population of immediate bracket responses provides a test for certain types of anchoring. ...
... For estimates of objective quantities as well as willingness to pay for public goods, they find that, for high entry point values, the proportion of bracket respondents who report that the amount is larger than the entry point value exceeds the fraction of open-ended answers larger than the entry point value. Hurd et al. (1998) specify a parametric model that captures this phenomenon in a symmetric way, biasing probabilities smaller than 0.5 upwards and probabilities larger than 0.5 downward. The idea is that respondents do not compare the true value Y to the entry point E, but instead compare Y to E+ε, where ε is a mean zero error term, assumed to be normal and independent of all other components of the model. ...
... The analysis draws on two data sources: the Health contains detailed questions about the financial situation of women of varying marital statuses. The HRS asks about multiple sources of income and assets and makes extensive use of unfolding brackets when collecting financial information, a technique shown to be highly effective (for example, Hurd et al., 1998). 3 In addition, we have adequate sample sizes to examine differences in income and wealth by marital status. ...
Social Security Administration http://deepblue.lib.umich.edu/bitstream/2027.42/50573/1/wp046.pdf
... Few surveys contain broad consumption questions; here we concentrate on the experience of the Canadian Out of Employment Panel (COEP) and the Italian Survey on Household Income and Wealth (SHIW). We also discuss pilot studies conducted in the US on the Asset and Health Dynamics among the Oldest Old (AHEAD) sample (Hurd et al.,1998) and in the Netherlands on the CentER Panel sample (Winter, 2001). The Canadian Out of Employment Panel (COEP) sampled about 20,000 Canadian workers who separated from a job in either 1993 or 1995. 2 In this survey a number of questions were asked concerning expenditures in the past week or month on a range of goods including housing; food at home; food outside the home; clothing and total expenditure. ...
... ty of the expenditure data in the 1995 wave has been assessed by Battistin, Miniaci and Weber (2003) (hereafter BMW) by comparing it to the corresponding diary based survey, the Survey of Family Budgets (SFB), run by the official statistical office (ISTAT). One issue that is often raised in this context is the item response rate for such questions. Hurd et al. (1998) record a relatively low response rate of 64% for their sample and go on to an unfolding brackets procedure to increase the response rate. Our findings are somewhat different.Table 1 provides some item non-response rates for the total expenditure and other questions in the COEP. The item response rate is high and not significantly worse ...
In many research areas it is desirable to have information on household total expenditure ('consumption'). We draw evidence from several sources on the usefulness of recall consumption questions. We conclude that valid information can be collected by adding specific recall questions to general purpose surveys, and provide recommendations on how to do so. Copyright 2003 Royal Economic Society.
