Public Opinion Quarterly

Public Opinion Quarterly 2006 70(5):704-719; doi:10.1093/poq/nfl032

This Article Abstract FREE Full Text (PDF) Supplementary Data Supplementary Data Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Johnson, T. P. Articles by Holbrook, A. L. Search for Related Content Social Bookmarking          What's this?

© The Author 2006. Published by Oxford University Press on behalf of the American Association for Public Opinion Research. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org.

Using Community-Level Correlates to Evaluate Nonresponse Effects in a Telephone Survey

Timothy P. Johnson, Young IK Cho, Richard T. Campbell and Allyson L. Holbrook

TIMOTHY P. JOHNSON, YOUNG IK CHO and ALLYSON L. HOLBROOK are with the Survey Research Laboratory, and richard t. campbell is with the Institute for Health Policy Research, all at the University of Illinois at Chicago.

Address correspondence to Timothy Johnson; e-mail: timj{at}uic.edu.

	Abstract

TOP Abstract Introduction Methods Findings Discussion Supplementary Data Appendix References

 
Understanding the relationship between nonresponse processes and key research variables is central to evaluating if and how nonresponse introduces bias into survey estimates. In most telephone surveys, however, little information is available with which to estimate these effects. We report a procedure for examining the potential effects of nonresponse via analyses that (1) investigate the linkages between community-level (zip code) variables and survey nonresponse and (2) examine the associations between these community-level variables and key survey measures. We demonstrate these procedures using hierarchical modeling to analyze data from a state-wide telephone survey in Illinois. One zip code–level indicator of concentrated disadvantage—the percentage of the population below poverty level—was found to be positively associated with nonresponse and, among respondents, with both current physical disability status and lack of health insurance coverage, suggesting that both may have been underestimated in this survey. This inexpensive approach has the potential of enabling researchers to routinely evaluate nonresponse effects in their survey data.

---

	Introduction

TOP Abstract Introduction Methods Findings Discussion Supplementary Data Appendix References

 
Response rates to social and behavioral surveys have been declining for several decades (de Leeuw and de Heer 2002; Groves and Couper 1998; Steeh 1981), increasing the likelihood that differences between respondents and nonrespondents may be sufficient to bias survey estimates of means, proportions, and other population parameters (Caetano 2001). Unit nonresponse is a form of missing data. Because it is unavoidable, such missing data should ideally be, in the terminology of Little and Rubin (2002), missing completely at random. That is, the probability of a response should preferably be independent of the true value of the item in question, were it to be available, and independent of all other variables, measured or unmeasured. Barring that, we would at least like nonresponse to be missing at random and thus ignorable, meaning that the probability of a response is unrelated to the true value of the variable in question within categories of the control variables. In a survey context, this would mean that after weighting adjustments for poststratification variables, and perhaps after some explicit model-based estimation, it is possible to arrive at an unbiased estimate of the population parameter in question. Unfortunately, it is never possible to ascertain with certainty that missing data are ignorable, but it is useful to search for variables potentially associated with nonresponse in an effort to adjust estimates to the maximum extent possible. And, to the extent that nonresponse can be shown to be unrelated to available information about respondents, some additional confidence regarding ignorability is warranted.

Several methodologies have been employed to investigate the potential effects of nonresponse on survey estimates. One is a two-phase survey, which attempts to conduct follow-up interviews with persons not responding to the initial survey (Gmel 2000; Hill et al. 1997; Lahaut et al. 2002). A second involves comparisons of early versus late responders to survey requests (Etter and Perneger 1997; Voigt, Koepsell, and Daling 2003). These approaches are limited by questions of how representative follow-up or late respondents are of all nonrespondents, and by the frequent use of multiple data collection modes that may confound nonresponse error with measurement quality.

Another potential approach to addressing nonresponse bias involves use of a record-matching strategy to identify variables that are available for both respondents and nonrespondents and are predictive of survey nonresponse. Some studies have employed record-matching to identify nonresponse correlates in face-to-face surveys (Gfroerer, Lessler, and Parsley 1997; Groves and Couper 1998; Needle, McCubbin, and Lorence 1985). Although useful, none of these studies extend this technique by taking the additional step of identifying the presence and direction of nonresponse effects in survey estimates. In this article we do so by investigating nonresponse bias in a statewide telephone survey. Identifying techniques for assessing nonresponse bias in telephone surveys is particularly important because very little information is typically available regarding nonresponding households for this mode of data collection.

	Methods

TOP Abstract Introduction Methods Findings Discussion Supplementary Data Appendix References

 
Between January 15 and August 15, 2003, 4,155 random telephone interviews regarding substance use treatment needs in the state of Illinois were completed by the University of Illinois at Chicago Survey Research Laboratory. The sample was stratified in order to produce reliable estimates for each of eight geographic areas of the state. Whenever possible, sampled households were sent an advance letter introducing the survey. The AAPOR response rate (RR3) for the study was 32.7 percent (AAPOR, 2000).

Within each sampled household, one person aged 16 years or older was randomly selected to be interviewed; for those aged 16 and 17, parental consent was obtained in advance. The parental approval rate for participation by their underage children was 68 percent. Interviews averaged 29.8 minutes in length (standard deviation = 12.7) and were primarily concerned with issues of substance use involvement and relevant risk factors. The study was reviewed, approved, and monitored by the University of Illinois at Chicago institutional review board.

To examine the potential effects of nonresponse error in this study, indicators of survey response, contact, and refusal were first constructed using the final sample dispositions assigned to each of the 20,774 telephone numbers included in the survey’s sample frame. The first indicator contrasted those households completing telephone interviews with all households identified as eligible that did not participate. A second measure contrasted eligible households where the selected respondent was never directly contacted with those where the selected respondent was contacted. The third indicator contrasted eligible households classified as refusals with nonrefusing eligible households. Households not speaking English or Spanish were excluded from these analyses. For complete disposition coding information, please see the supplementary material online.

Data from several other sources were next merged with this sample frame file for analysis. First, survey responses for six substantive measures of interest were appended to those phone numbers that yielded completed interviews: (1) past year alcohol or drug abuse treatment need; (2) lack of current health insurance coverage; (3) current physical disability; (4) ever having been homeless; (5) past year receipt of public aid; and (6) lifetime problem gambling. These measures varied by time frame, with two each representing lifetime, past year, and current respondent conditions and experiences. Four of these estimated measures were based on responses to multiple survey questions (see the appendix for the wording of all survey items). Past year alcohol or drug abuse treatment need was measured using Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) criteria for substance abuse or dependence (American Psychiatric Association 1994). Lifetime problem gambling was also measured using DSM-IV diagnostics (Gerstein et al. 1999). Current physical disability was measured using items adapted from the long form (questions 16a and b) of the 2000 U.S. Census questionnaire (Bureau of the Census 2000). Lifetime homeless experience was measured using an item first employed by Link et al. (1994). Past year receipt of public aid was based on questions regarding household receipt of income from unemployment compensation, workers’ compensation, Supplemental Security Income (SSI), Veterans Administration (VA), or Temporary Assistance to Needy Families (TANF) during the past 12 months. Lack of current health insurance was measured via a single question that asked, "Do you currently have health insurance coverage?"

Next, each telephone exchange/area code combination included in the survey’s sample frame was linked with the zip code for its central switching station using a database supplied by Melissa Data (2001). Data were available for a sample of 866 zip codes in Illinois. Because zip codes and telephone exchanges do not overlap perfectly, we evaluated the degree of error that was introduced by our matching process. We compared respondents’ reports of their zip codes with the zip code matching data and found a perfect match for only 60 percent of the cases. Because zip codes sharing the same first three digits are geographically contiguous, we also examined the degree of matching for these three digits only and found a 97 percent match rate.

Six measures from the 2000 census, obtained via the American FactFinder Web site (http://factfinder.census.gov) and aggregated at the zip code level, were next merged with the sample frame information. Of course, numerous census-based measures are available. Collinearity considerations prevent all of them from being used, and data reduction techniques based on factor or cluster analysis present their own problems. Instead, we chose measures previously identified by Sampson and colleagues (Sampson, Morenoff, and Earls 1999; Sampson, Raudenbush and Earls 1997) as predictors of collective efficacy, a construct they define as the willingness of citizens "to intervene on behalf of the common good" (Sampson, Raudenbush, and Earls 1997, p. 918). The specific measures included were (1) percentage of population at the same address for past 5 years (an indicator of residential stability); (2) percentage of adults with professional or managerial occupations (an indicator of concentrated affluence); (3) percentage of population that is below poverty level (an indicator of concentrated disadvantage); (4) population density (an indicator of urbanicity); (5) percentage of population that is foreign-born (an indicator of concentrated immigration); and (6) the adult-to-child ratio (an indicator of emphasis on children and family). Because participation in social surveys has also been interpreted as a form of community involvement (Couper, Singer, and Kulka 1998), examining the effects of these constructs on survey response propensity introduces a potentially useful new theoretical framework for assessing survey nonresponse.

Using unweighted survey data, hierarchical models were estimated in HLM6 to examine potential associations between each of the zip code measures and both household response measures and the six substantive variables described above (Raudenbush and Bryk 2002; Raudenbush et al. 2004). Since each outcome of interest was binary (coded "0" for no and "1" for yes), the first-level equations used a logit link function (Hedeker and Gibbons 1994) to estimate the log-odds of a positive response. All estimates presented are population-averaged with robust standard errors. Because the sample was stratified by state geographic region, this variable was included as a fixed effect covariate in all models. Models that examine substantive measures additionally include several demographic covariates, including gender, age, and race/ethnicity. In these analyses all predictor variables were mean-centered. The average probabilities of substantive measures, obtained from the intercepts of these models, were subsequently compared with the probabilities from models without the zip code–level variables. Z-statistics were used to test for significant differences between these models.¹

	Findings

TOP Abstract Introduction Methods Findings Discussion Supplementary Data Appendix References

 
SURVEY PARTICIPATION
The HLM results presented in table 1 examine the zip code–level predictors of three measures of survey response propensity. The first model (equation 1) examines associations with the survey response measure. Three variables were found to be negatively associated with likelihood of survey response: concentrated affluence (percentage of adults with professional or managerial occupations); concentrated disadvantage (percentage of population below the poverty level); and urbanicity (population density). These findings indicate that households in areas with higher concentrations of affluence were less likely to participate in the survey, as were households in disadvantaged areas and those in more urbanized areas in general. The second equation, for survey noncontact, also identifies concentrated affluence as being positively associated with survey noncontact. That is, eligible respondents within households in more affluent areas were less likely to be successfully contacted for this survey. The final equation, for survey refusal (equation 3), found two measures to be positively associated with survey refusal among eligible households: concentrated affluence and residential stability.

View this table: [in this window] [in a new window]   Table 1. Hierarchical Model of Effects of Zip Code–Level Variables on Survey Response, Contact, and Refusals

 
SURVEY OUTCOMES
Six additional HLM models were specified to investigate the effects of these zip code–level measures on each substantive survey measure of interest, for those respondents who were interviewed. Table 2 presents these models. As shown in equation 4, when controlling for individual-level demographic measures, none of the zip code–level measures were found to be independently associated with past year alcohol and drug use. However, the percentage of the population below poverty level (concentrated disadvantage) and percentage of population that is foreign-born (concentrated immigration) were both negatively associated with currently having health insurance (equation 5), and percentage below poverty level was positively associated with currently having a physical disability (equation 6).

View this table: [in this window] [in a new window]   Table 2. Hierarchical Models of Joint Effects of Zip Code– and Individual-Level Variables on Substantive Measures

 
A model examining the variables associated with ever having been homeless is presented in equation 7. None of the six zip code–level measures were found to be independently associated with this measure. In equation 8 a similar model is constructed to examine receipt of public aid during the past year. In this model one zip code–level measure, percentage of the population living at the same address for the past 5 years (residential stability) was found to be positively associated with receipt of public aid. As shown in equation 9, having a lifetime gambling problem was positively associated with the adult-to-child ratio within each zip code. That is, persons living in areas with smaller proportions of children (i.e., low emphasis on children and family) were at greater risk for having a gambling problem.

In addition, race/ethnicity was found to be associated with all six of these substantive measures; age was associated with five, and gender was associated with three. Specifically, African-American respondents were more likely to report (relative to whites) having a current physical disability, having ever been homeless, having received public aid in the past year, to be medically uninsured, and having a lifetime gambling problem, and they were less likely than whites to report past year alcohol or drug abuse. Latinos were more likely than whites to report having ever been homeless and to have no health insurance, and they were less likely to report a lifetime gambling problem or past year alcohol or drug abuse. Persons of other race/ethnicities were also less likely than whites to report past year alcohol or drug abuse. Younger respondents were at increased risk for not currently having health insurance, for reporting past year alcohol or drug abuse or having received public aid in the past year, and for reporting a lifetime gambling problem. Older respondents were at increased risk for having a current physical disability. Males were at greater risk for past year alcohol or drug abuse and a lifetime gambling problem, and they were less likely to be medically uninsured.

Findings in table 2 indicate that two zip code measures were associated both with the likelihood of survey response among sampled households and with two of the survey’s substantive measures among participating households. The direction of these relationships indicates that persons living in areas of concentrated disadvantage were less likely to be included in the survey, and that respondents within these areas were more likely to be medically uninsured and to have a physical disability. These results suggest that survey estimates of the population without health insurance coverage may be underestimated, because persons living in zip codes with higher concentrations of poverty were less likely to participate, and respondents in these areas were less likely to report having health insurance. Likewise, the prevalence of physical disabilities may also be underestimated, given higher nonresponse within zip codes with concentrated poverty and higher reporting of physical disabilities among respondents within these areas.

These conclusions were reinforced by comparisons of prevalence estimates from HLM models that do and do not include adjustments for the zip code–level characteristics. When adjustments for the six zip code measures were introduced, the percentage of the sample reporting no current health insurance increased from 9.0 percent (95 percent confidence intervals = 8.1–10.1) to 11.1 percent (confidence intervals = 8.1–12.9). The estimated percentage with a current physical disability increased from 18.8 percent (confidence intervals = 17.4–20.2) to 20.5 percent (confidence intervals = 17.7–23.7) when a similar adjustment for these zip code measures was introduced.

	Discussion

TOP Abstract Introduction Methods Findings Discussion Supplementary Data Appendix References

 
These findings support the work of Sampson and colleagues (Sampson, Morenoff, and Earls 1999; Sampson, Raudenbush, and Earls 1997) by linking three of their community-level predictors of collective efficacy to survey response, which other work on survey participation suggests may be a related construct (Couper, Singer, and Kulka 1998). Specifically, households within areas of concentrated affluence, those within areas of concentrated disadvantage, and those within highly urbanized areas were less likely to participate in a random digit dial (RDD) telephone survey. Concentrated affluence was also predictive of noncontact and refusal. Urbanicity and higher household socioeconomic status have both been linked to greater nonresponse in previous studies (Groves and Couper 1998). An association between higher community disadvantage and survey nonresponse, however, has not to our knowledge been previously reported. Other work (Massey and Denton 1993; Swaroop and Morenoff 2006; Wilson 1987), though, suggests that neighborhoods plagued with concentrated poverty also experience considerable social disorganization, which may lead to distrust and may also suppress survey participation. Also of note was the finding of higher refusal rates in areas with less residential mobility. Although somewhat counterintuitive, similar results have been reported previously (Comstock and Helsing 1973; Groves and Couper 1998).

We also found that one of these measures, an indicator of concentrated disadvantage, was predictive of two of the substantive variables estimated from respondents in this survey: lack of health insurance and current physical disability. These effects were independent of respondent-level demographic measures (i.e., gender, age, and race/ethnicity) and geographic stratification measures (i.e., Illinois region). The direction of these effects suggests that the survey may have underestimated the proportion of respondents currently without health insurance in Illinois, as well as the proportion with physical disabilities—findings that were supported by model-based comparisons of estimates developed with and without the introduction of these zip code indicators. It is noteworthy that these nonresponse effects were observed for the two measures of current respondent conditions but not for those assessing longer (i.e., lifetime or past year) time frames. Although this merits further investigation, we speculate that population mobility patterns may attenuate associations between zip code–level variables and more distant respondent experiences.

We believe this technique may be a useful approach for researchers concerned with the possible implications of nonresponse error for survey estimates. Although past research has examined community-level correlates of nonresponse (Gfroerer, Lessler, and Parsley 1997; Groves and Couper 1998; Needle, McCubbin, and Lorence 1985), our approach goes further to estimate the effect of nonresponse on specific survey measures. By identifying variables that are correlated with nonresponse among all sampled households and with substantive measures of interest among respondents, researchers can more directly estimate and adjust for nonresponse error. Conversely, to the extent that substantive measures are not associated with community-level predictors of nonresponse, researchers can have some additional confidence that nonresponse error is not seriously biasing estimates. Although this methodology is certainly applicable to face-to-face surveys, we believe it is particularly valuable in RDD studies, given the relative lack of other available strategies for assessing nonresponse error when collecting data via telephone.

One of the advantages of the approach used in this article is that it could routinely be used in telephone surveys to estimate and adjust for nonresponse bias. This is important because survey participation may vary as a result of respondents’ level of interest in the survey topic (e.g., Groves, Presser, and Dipko 2004) and for other reasons (Groves, Singer, and Corning 2000). As a result, the correlates of nonresponse may vary from survey to survey and may be influenced by the characteristics of the introduction to the survey or information about the survey provided to a respondent through an advance letter or answering machine message. If the correlates of nonresponse vary from survey to survey, developing a generalizable model of survey nonresponse may not be possible. Instead, researchers would need a tool, such as the one described in this article, to estimate and adjust for nonresponse in each survey.

In evaluating this approach, we emphasize that assessment of the ecological correlates shared by response propensity and substantive measures can only be conducted among survey respondents. This strategy thus provides only indirect evidence that response correlates are associated with outcome measures for nonrespondents. It should also be noted that the total population size varies considerably by zip code, ranging from 169 to 113,986, with a mean of 9,963 (SD = 16,371). Consequently, the size of the geographic areas being compared and the degree of heterogeneity within each cannot be considered uniform, and this variability may attenuate the relationships being examined (Krieger et al. 2002). For these reasons, as well as concerns about the imperfect matching of phone numbers with zip codes, it is unclear that the zip code level of aggregation is most appropriate for analyses such as those reported here. This technique may also become less effective in the future, as telephone numbers become less geographically fixed. Nonetheless, it provides a relatively inexpensive and accessible method for investigating the ecological correlates of nonresponse and for linking potential nonresponse correlates with critical survey measures in RDD surveys. As such, this methodology warrants further investigation.

	Supplementary Data

TOP Abstract Introduction Methods Findings Discussion Supplementary Data Appendix References

 
Supplementary data are available online at http://pubopq.oxfordjournals.org/.

	Appendix

TOP Abstract Introduction Methods Findings Discussion Supplementary Data Appendix References

 
WORDING OF SURVEY QUESTIONS INCLUDED IN ANALYSIS
Past Year Alcohol or Drug Abuse
The following questions are repeated for marijuana, cocaine/crack, heroin, pain relievers/opiates, methamphetamines and stimulants, hallucinogens, tranquilizers and sedatives, inhalants, club drugs, and other drugs. Response options are Yes, No, No coded response applicable, Don’t know/Refused.

1. Now I am going to ask some more questions about your [alcohol] use in the past year. Was there a time when you spent a lot of time using [alcohol], getting over its effects, or obtaining it?

2. Was there a time when you used [alcohol] more often or in larger amounts than you intended to?

3. Was there a time when using the same amount of [alcohol] had less effect than before, or it took more to feel the same effect?

4. Was there a time when your use of [alcohol] often kept you from working, going to school, taking care of children, or taking part in recreational activities?

5a. Was there a time when your use of [alcohol] caused you to have emotional or psychological problems such as feeling uninterested in things, depressed, suspicious of people, or paranoid?

5b. Did you continue to use [alcohol] in spite of this?

6. Was there a time when your use of [alcohol] caused you to have any physical health problems?

6a. Did you continue to use [alcohol] in spite of this?

7. Was there a time when you wanted to stop using or cut down on [alcohol], but found that you couldn’t?

8. Was there a time when you made rules about where, when, or how much you would use [alcohol], and then broke the rules?

9a. Was there a time when you experienced anxiety, sweating, hands trembling, or heart beating fast as the effect of [alcohol] was wearing off?

9b. Was there a time when you had trouble sleeping or had bad dreams as the effect of [alcohol] was wearing off?

9c. Was there a time when you vomited or felt nauseous as the effect of [alcohol] was wearing off?

9d. Was there a time when you experienced seeing, hearing, or feeling things that weren’t really there as the effect of [alcohol] was wearing off?

9e. Was there a time when you felt either very slowed down or like you couldn’t sit still as the effect of [alcohol] was wearing off?

9f. Was there a time when you experienced seizures or fits as the effect of [alcohol] was wearing off?

9g. Was there a time when you felt exhausted, or slept more than you usually do, as the effect of [alcohol] was wearing off?

9h. Was there a time when you experienced diarrhea as the effect of [alcohol] was wearing off?

9i. Was there a time when you experienced cramps or muscle aches as the effect of [alcohol] was wearing off?

9j. Was there a time when you experienced eating either more or less than you usually do as the effect of [alcohol] was wearing off?

10. Was there a time when you drank [alcohol] to prevent or cure these problems?

Currently Has Health Insurance

Do you currently have health insurance coverage?

Currently Has a Physical Disability

1. Do you have any of the following long-lasting conditions:
   a. Blindness, deafness, or a severe vision or hearing impairment.
   b. A condition that substantially limits one or more basic physical activities (e.g., walking, climbing stairs, reaching, lifting, carrying).
   
2. Because of a physical, mental, or emotional condition lasting 6 months or more, do you have any difficulty in doing any of the following activities:
   c. Any difficulty learning, remembering, or concentrating.
   d. Any difficulty dressing, bathing, or getting around inside the home.
   e. Any difficulty going outside the home alone to shop or visit a doctor’s office.
   f. Any difficulty working at a job or business.
   

Ever Homeless
Have you ever had a time in your life when you considered yourself homeless?

Received Public Aid in Past Year
In the past 12 months, have you received any income or money from the government?

[IF YES]: Did you receive income:

1. From Supplemental Security Income or SSI for which you qualify because of a disability?
   
2. From Social Security or other retirement benefits you, your spouse, or your parents earned through work?
   
3. From Veterans Administration payments?
   
4. From unemployment compensation because of layoff, or workers’ compensation because of injuries at work?
   
5. From Temporary Assistance to Needy Families, TANF?
   
6. From other forms of public assistance such as Food Stamps?
   
7. From a social service agency or program? (SPECIFY)
   

Lifetime Gambling Problem
Please indicate which of the following types of gambling you have done in your lifetime:

1. Played cards for money (not at all, less than once a week, once a week or more)
   
2. Bet on horses, dogs, or other animals (in off-track betting, at the track, or with a bookie)
   
3. Played dice games for money
   
4. Gambled in a casino (legal or otherwise)
   
5. Played the numbers or bet on lotteries
   
6. Played bingo for money
   
7. Played the stock and/or commodities market
   
8. Played slot machines, poker machines, or other gambling machines
   
9. Bowled, shot pool, played golf, or played some other games of skill for money
   
10. Played pull tabs or "paper" games other than lotteries
    
11. Bet on some form of gambling not listed above
    

[IF INDICATED ANY OF THE ABOVE]

1. What is the largest amount of money you have ever gambled with on any one day?
   
2. Do (did) your parents have a gambling problem? (both my father and mother gamble[d] too much; my father gamble[d] too much; my mother gamble[d] too much; neither one gamble[d] too much)
   
3. When you gamble, how often do you go back another day to win back money you lost? (never; some of the time I lost; most of the time I lost; every time I lost)
   
4. Have you ever claimed to be winning money gambling but weren’t really—in fact, you lost? (never; yes, less than half the time I lost; yes, most of the time I lost)
   
5. Do you feel you have ever had a problem with gambling? (yes; yes—in the past but not now; no)
   
6. Have people criticized your betting or told you that you had a gambling problem, regardless of whether or not you thought it was true?
   
7. Have you ever felt guilty about the way you gamble or what happens when you gamble?
   
8. Have you ever felt like you would like to stop betting money or gambling but didn’t think you could?
   
9. Have you ever hidden betting slips, lottery tickets, gambling money, IOUs, or other signs of betting or gambling from your spouse, children, or other important people in your life?
   
10. Have you ever argued with people you live with over how you handle money?
    
11. (IF YES TO j): Have money arguments ever centered on your gambling?
    
12. Have you ever borrowed from someone and not paid them back as a result of your gambling?
    
13. Have you ever lost time from work (or school) due to betting money or gambling?
    
14. If you borrowed money to gamble or to pay gambling debts, whom or where did you borrow from?
    

1. From household money?
   
2. From your spouse?
   
3. From other relatives or in-laws?
   
4. From banks, loan companies, or credit unions?
   
5. From credit cards?
   
6. From loan sharks?
   
7. You cashed in stocks, bonds, or other securities?
   
8. You sold personal or family property?
   
9. You borrowed on your checking account (passed bad checks)?
   
10. You have (had) a credit line with a bookie?
    
11. You have (had) a credit line with a casino?
    

The specific algorithms used to construct final measures of past year alcohol or drug abuse, current physical disability, receipt of public aid in past year, and lifetime gambling problem using these survey questions are available from the authors.

	Footnotes

 
An earlier version of this article was presented at the annual meeting of the American Association for Public Opinion Research, Phoenix, AZ, May 14, 2004.

^1. The Z-statistic formula used to compare two probabilities is Z = (P₁–P₀)/SQRT{[P₀(1–P₀)]/n}, where P₁ = Probability of outcome based on the equations with zip code–level variables, P₀ = Probability of outcome from equations without zip code–level predictors, and n = sample size.

	References

TOP Abstract Introduction Methods Findings Discussion Supplementary Data Appendix References

 
American Psychiatric Association. (1994) Diagnostic and Statistical Manual of Mental Disorders 4th ed. (DSM-IV) (American Psychiatric Association, Washington, DC).

American Association for Public Opinion Research (AAPOR). (2000) Standard Definitions: Final Dispositions of Case Codes and Outcome Rates for Surveys(AAPOR, Ann Arbor, MI).

Bureau of the Census. (2000) United States Census 2000: Form D-2(Department of Commerce, Washington, DC) Available online at http://www.census.gov/dmd/www/pdf/d02p.pdf (accessed April 2, 2006).

Caetano Raul. (2001) Nonresponse in Alcohol and Drug Surveys: A Research Topic in Need of Further Attention. Addiction 96:1541–45.[CrossRef][Web of Science][Medline]

Comstock George W and Helsing. Knud J. (1973) Characteristics of Respondents and Nonrespondents to a Questionnaire for Estimating Community Mood. American Journal of Epidemiology 97:233–39.[Web of Science][Medline]

Couper Mick P, Singer Eleanor, Kulka Richard A. (1998) Participation in the 1990 Decennial Census: Politics, Privacy, Pressures. American Politics Quarterly 26:59–80.

de Leeuw Edith and de Heer Wim. (2002) Trends in Household Survey Nonresponse: A Longitudinal and International Comparison. In Groves Robert M, Dillman Don A, Eltinge John L, Little Roderick J. A (Eds.). Survey Nonresponse(Wiley, New York) pp. 41–54.

Etter Jean-François and Thomas V. Perneger. (1997) Analysis of Nonresponse Bias in a Mailed Health Survey. Journal of Clinical Epidemiology 50:1123–28.[CrossRef][Web of Science][Medline]

Gerstein Dean, Murphy Sally, Toce Marianna, Hoffman John, Palmer Amanda, Johnson Robert, et al. (1999) Gambling Impact and Behavior Study: Report to the National Gambling Impact Study Commission(National Opinion Research Center, Chicago) Available online at http://www.norc.uchicago.edu/new/pdf/gamble.pdf (accessed April 2, 2006).

Gfroerer Joseph, Lessler Judith, Parsley Teresa. (1997) Studies of Nonresponse and Measurement Error in the National Household Survey on Drug Abuse. In Harrison Lana and Hughes rthur (Eds.). The Validity of Self-Reported Drug Use: Improving the Accuracy of Survey Estimates(National Institute on Drug Abuse, Rockville, MD) pp. 273–95.

Gmel Gerhard. (2000) The Effect of Mode of Data Collection and of Nonresponse on Reported Alcohol Consumption: A Split-Sample Study in Switzerland. Addiction 95:123–34.[CrossRef][Web of Science][Medline]

Groves Robert M and Couper Mick P. (1998) Nonresponse in Household Interview Surveys(John Wiley and Sons, New York).

Groves Robert M, Presser Stanley, Dipko Sarah. (2004) The Role of Topic Interest in Survey Participation Decisions. Public Opinion Quarterly 68:2–31.[Abstract/Free Full Text]

Groves Robert M, Singer Eleanor, Corning Amy D. (2000) Leverage-Saliency Theory of Survey Participation: Description and an Illustration. Public Opinion Quarterly 64:299–308.[CrossRef][Web of Science][Medline]

Hedeker Donald and Gibbons Robert. (1994) A Random-Effects Ordinal Regression Model for Multilevel Analysis. Biometrics 50:993–94.

Hill Anthony, Roberts Julian, Ewings Paul, Gunnell David. (1997) Nonresponse Bias in a Lifestyle Survey. Journal of Public Health Medicine 19:203–7.[Abstract/Free Full Text]

Krieger Nancy, Waterman Pamela, Chen Jarvic T, Soobader Mah-Jabeen, Subramanian S. V, Carson Rosa. (2002) Zip Code Caveat: Bias Due to Spatiotemporal Mismatches between Zip Codes and US Census–Defined Geographic Areas—The Public Health Disparities Geocoding Project. American Journal of Public Health 92:1100–1102.[Free Full Text]

Lahaut Viviënne M. H., Harrie A. M. Jansen, Dike van de Mheen, Henk F. L. Garretsen. (2002) Nonresponse in a Sample Survey on Alcohol Consumption. Alcohol and Alcoholism 37:256–60.[Abstract/Free Full Text]

Link Bruce G, Susser Ezra, Stueve Ann, Phelan Jo, Moore Robert E, Struening Elmer. (1994) Lifetime and Five-Year Prevalence of Homelessness in the United States. American Journal of Public Health 84:1907–12.[Abstract/Free Full Text]

Little Roderick and Rubin Donald. (2002) Statistical Analysis with Missing Data(John Wiley and Sons, Hoboken, NJ).

Massey Douglas and Denton Nancy. (1993) American Apartheid: Segregation and the Making of the Underclass(Harvard University Press, Cambridge, MA).

Melissa Data. (2001) Fone*Data User’s Guide(Melissa Data Corporation, Rancho Santa Margarita, CA).

Needle Richard, McCubbin Hamilton, Lorence Jon. (1985) A Test of Non-Respondent Bias in a Family-Based Study: A Research Note. International Journal of the Addictions 20:763–69.[Web of Science][Medline]

Raudenbush Stephen and Bryk Anthony. (2002) Hierarchical Linear Models: Applications and Data Analysis Methods(Sage, Thousand Oaks, CA).

Raudenbush Stephen W, Bryk Anthony S, Cheong Yuk F, Congdon Richard T. (2004) Hierarchical Linear and Nonlinear Modeling. HLM 6 (Scientific Software International, Lincolnwood, IL).

Sampson Robert J, Morenoff Jeffrey D, Earls Felton. (1999) Beyond Social Capital: Spatial Dynamics of Collective Efficacy for Children. American Sociological Review 64:633–60.[CrossRef]

Sampson Robert J, Raudenbush Stephen, Earls Felton. (1997) Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy. Science 277:918–24.[Abstract/Free Full Text]

Steeh Charlotte G. (1981) Trends in Nonresponse Rates, 1952–1979. Public Opinion Quarterly 45:40–57.[CrossRef][Web of Science]

Swaroop Sapna and Morenoff Jeffrey D. (2006) Building Community: The Neighborhood Context of Social Organization. Social Forces 84:1665–96.

Voigt Linda F, Koepsell Thomas D, Daling Janet R. (2003) Characteristics of Telephone Survey Respondents According to Willingness to Participate. American Journal of Epidemiology 157:66–73.[Abstract/Free Full Text]

Wilson William J. (1987) The Truly Disadvantaged: The Inner City, the Underclass, and Public Policy(University of Chicago Press, Chicago).

CiteULike    Connotea    Del.icio.us    What's this?

This Article Abstract FREE Full Text (PDF) Supplementary Data Supplementary Data Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Johnson, T. P. Articles by Holbrook, A. L. Search for Related Content Social Bookmarking          What's this?

Online ISSN 1537-5331 - Print ISSN 0033-362X

Copyright © 2009 American Association for Public Opinion Research

Oxford Journals Oxford University Press

• Site Map
• Privacy Policy
• Frequently Asked Questions

Other Oxford University Press sites: Oxford University Press Oxford Journals China Oxford Journals Japan American National Biography Booksellers' Information Service Children's Fiction and Poetry Children's Reference Corporate & Special Sales Dictionaries Dictionary of National Biography Digital Reference English Language Teaching Higher Education Textbooks Humanities International Education Unit Law Medicine Music Online Products Oxford English Dictionary Reference Rights and Permissions Science School Books Social Sciences Very Short Introductions World's Classics