Public Opinion Quarterly Advance Access originally published online on November 13, 2007
Public Opinion Quarterly 2007 71(5):734-749; doi:10.1093/poq/nfm047
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||
Published by Oxford University Press 2007.
Coverage Bias in Traditional Telephone Surveys of Low-Income and Young Adults
Address correspondence to Stephen J. Blumberg; e-mail: sblumberg{at}cdc.gov
 |
Abstract |
|---|
 TOP Abstract IntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
The proportion of adults with only wireless telephones is growing rapidly. Using 2006 data from the National Center for Health Statistics’ National Health Interview Survey, this article is among the first to reveal that noncoverage of this population can result in nonnegligible bias for traditional random-digit-dial landline telephone surveys that do not call wireless telephone numbers. In 2006 in the United States, 17 percent of low-income adults with household income below 200 percent of the federal poverty thresholds, 25 percent of young adults aged 18–29 years, and 32 percent of low-income young adults lived in households with only wireless telephones. Within each of these three subgroups, we compared wireless-only adults and adults with landline telephones on demographic characteristics and 13 key indicators of health status, health behaviors, health care service use, and health care access. Even after statistical adjustments that account for demographic differences between adults living in households with and without landlines, telephone surveys of landlines will underestimate the prevalence of health behaviors, such as binge drinking, smoking, and HIV testing. Obesity may be overestimated and physical activity may be underestimated for low-income young adults. No significant bias is predicted for other measures of health status and health insurance coverage. Sample weighting procedures that incorporate adjustments for multiple demographic characteristics are necessary to help attenuate coverage bias in traditional telephone surveys, but may not be sufficient for behavioral risk factor surveys of low-income and young adults.
|
Introduction |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
Estimates based on data from the 2006 National Health Interview Survey (NHIS) suggest that at least 1 in 8 American adults live in households with only wireless telephones, and this proportion is growing rapidly (Blumberg and Luke 2007
). Relative to adults living in households with landline telephones, these wireless-only adults are more likely to be young, living in low-income households, renting their homes, and living alone or with roommates. As might be expected for a group that includes a disproportionate number of young poor adults living apart from family, wireless-only adults are more likely to binge drink alcohol and smoke, and they are less likely to have health insurance coverage and a usual place for medical care. Yet, they are also less likely to be obese, less likely to have been diagnosed with diabetes, and more likely to have been tested for HIV (Blumberg, Luke, and Cynamon 2006; Blumberg and Luke 2007). Survey researchers remain concerned about the potential implications of such differences for traditional random-digit-dial (RDD) landline telephone surveys. Most survey research organizations exclude wireless telephone numbers from their RDD sample frames. Therefore, these young poor adults are less likely to be contacted to participate in RDD landline surveys, relative to other adults. This undercoverage can potentially lead to biased survey estimates, though research based on NHIS data has suggested that the magnitude of this potential bias is generally less than 2 percentage points for estimates of health care service use and health status of all adults (Blumberg et al., forthcoming). Differences this small or smaller have been described as "not practically significant" (Blumberg, Luke, and Cynamon 2006), and thus may be regarded as "negligible."
The significance of differences for specific subgroups of the population – such as young adults or adults with low income – is unknown. Estimates for such subgroups, and comparisons of one subgroup to other subgroups, are often a greater focus of surveys than are estimates for the total population. Because biases for subgroups of the population may not be adequately reflected in observed biases for the overall population, we sought to determine whether the exclusion of adults without landlines may bias subgroup estimates derived from health-related telephone surveys. We selected three subgroups based on their low rates of landline telephone ownership and their importance to health policymakers: (1) young adults aged 18–29 years, (2) adults living in households with low income (hereafter referred to as "low-income adults"), and (3) the intersection of these two subgroups (low-income young adults).
|
Data Source |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
Estimates presented here are derived from data collected by the 2006 National Health Interview Survey. This cross-sectional, in-person, household survey of the US civilian noninstitutionalized population, conducted annually by the National Center for Health Statistics of the Centers for Disease Control and Prevention, is designed to collect information on health status, health-related behaviors, and health care utilization (Botman et al. 2000). The survey also includes questions about household telephones and whether "you or anyone in your family has a working cellular telephone." A family can be an individual or a group of two or more related persons living together in the same housing unit. More than one family can live in a household (including, for example, a household where there are multiple single-person families, as when unrelated roommates are living together).
In this article, a family is identified as a "wireless family" if anyone in the family had a working cellular telephone. A household is identified as "wireless-only" if there are no working landline telephones inside the household and if the household includes at least one wireless family. Adults are identified as wireless-only if they live in a wireless-only household. A similar approach is used to identify adults living in households with no telephone service (neither wireless nor landline, hereafter referred to as "phoneless").
Household telephone status (rather than family telephone status) is used in this report because most telephone surveys of the general public draw samples of households rather than families. Individual telephone ownership is not considered, in large part because such information is not available from the NHIS. It is possible that an individual living in a household with landline telephone service may consider himself or herself to be wireless-only if the individual does not use the landline telephone. Similarly, an individual who does not own a wireless telephone may consider herself or himself to have no telephone service despite living with a wireless-only roommate. Nevertheless, the former individual is likely to be included in the household when a landline telephone survey asks about all persons living in the household, and the latter individual is likely to be included in the household if a wireless telephone survey asks about all persons living with the wireless telephone user.
In the 2006 NHIS, household telephone status information was obtained for 29,065 households. Analyses of demographic characteristics are based on data from the NHIS Family file. Data for 54,315 civilian adults aged 18 years and over and living in interviewed households were used in these analyses. The family-level response rate was 87.0 percent, calculated as the product of the percentage of eligible families who responded to the survey (99.6 percent) and the percentage of eligible households who responded to the survey (87.3 percent).
Analyses of selected health measures are based on data from the NHIS Sample Adult file. This file includes data for one civilian adult randomly selected from each family.1 In 2006, data on selected health measures were collected from 24,275 randomly selected adults. The sample adult response rate was 70.8 percent, calculated as the product of the family-level response rate and the percentage of adults identified as eligible who responded to the survey (81.4 percent).
|
Analysis Procedures |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
The identification of low-income adults was based on household poverty status, which is derived from household income and household size using the US Census Bureau's poverty thresholds (US Census Bureau 2007
). "Low-income adults" were defined as those living in households with incomes less than 200 percent of the poverty threshold. In 2006, 200 percent of the poverty threshold was equivalent to approximately $21,000 for a single person or $41,000 for a household of four. For households with multiple families, household income and household size were calculated as the sum of the multiple measures of family income and family size. Therefore, household income may not reflect income available for use by the individual, such as when unrelated roommates are living together or a young adult is living with his/her parents. Household income was based only on reported income. Where income was missing (approximately 30 percent of NHIS records), imputed income values were not available. Household telephone status for all adults was examined by age and income. Wireless telephone status is unknown for 11 percent of NHIS households because of interview break-offs. (The questions were asked at the end of the lengthy survey.) Wireless telephone status data were not imputed. Therefore, the percentages for wireless-only adults are likely to be underestimated by a small margin.
Next, within each of the three populations of interest for this report, we compared the demographic composition of wireless-only persons with the demographic composition of persons living in households with landlines. A similar comparison was then made between wireless-only persons and persons with landlines on 13 key indicators of health status, health behaviors, health care service use, and health care access. These key indicators were selected from among the key adult health indicators published quarterly by the NHIS Early Release Program (Schiller, Martinez, and Barnes 2006). We used the same specifications for these indicators as have been published elsewhere (Schiller, Martinez, and Barnes 2006; Blumberg and Luke 2007).
The potential bias in landline telephone surveys that exclude wireless telephones and phoneless households was then calculated as the estimated value for adults living in landline telephone households minus the "true" population value. Statistical significance tests to determine if this bias was greater than zero were calculated using a standard error of the difference term that accounted for nonindependence of the two groups (adults living in landline telephone households and all adults) by incorporating their covariance. This statistical test is equivalent to a direct comparison of persons living in households with landlines with persons living in households without landlines (a group that includes wireless-only persons and persons living in phoneless households).
For most major RDD surveys, sampling weights for the households with landline telephones are adjusted to match estimates of the demographic composition of the population with and without landlines from an independent source (e.g., Census Bureau). This adjustment reduces some of the bias that may result from noncoverage of the nonlandline population (e.g., by increasing the value of the weights for younger adults who are generally underrepresented in RDD surveys). The estimated values for all adults were calculated using NHIS sampling weights that are adjusted to Census control totals for sex, age, and race/ethnicity. The estimated values for adults living in households with landline telephones were calculated using the same weights, without further adjustments. Because these weights do not match estimates of the demographic composition of the population with and without landlines, the unadjusted bias estimate is likely to overestimate the actual bias that might be observed from a telephone survey whose weights are adjusted to Census control totals.
To explore whether adjustments for demographic characteristics would greatly reduce the potential coverage bias in health surveys, we calculated an adjusted bias estimate that accounted for group differences between adults living in landline households and all adults. Each key health indicator was regressed on household telephone status (landline or no landline) and 8 or 9 demographic characteristics (age, household poverty status, race/ethnicity, sex, education, employment status, household structure, geographic region, urban/rural status, and home ownership). Age was not included in logistic regression analyses for young adults and household poverty status was not included in logistic regression analyses for low-income adults. For each health indicator, an adjusted value for adults living in landline households was then calculated as the predictive margin (Graubard and Korn 1999) when household telephone status was "landline." The adjusted bias in landline telephone surveys was then calculated by subtracting the "true" 2006 NHIS population value from the predictive margin. As before, the statistical significance tests to determine if this bias was greater than zero were calculated using a standard error of the difference term that accounted for nonindependence of the two groups by incorporating their covariance.
Point estimates and standard errors were produced using SUDAAN software to account for the complex sample design of NHIS. Predictive margins were calculated using the PREDMARG statement in SUDAAN. Significant differences were identified using two-sided significance tests at the 0.05 level.
|
Prevalence and Demographic Characteristics |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
In 2006, one-in-four young adults (24.7 percent) lived in wireless-only households (see table 1). These young adults account for half of all wireless-only adults (50.9 percent). One-in-six low-income adults lived in wireless-only households, accounting for 41.0 percent of all wireless-only adults. One-in-three low-income young adults (32.4 percent) lived in wireless-only households and compose approximately one-fourth (23.8 percent) of all wireless-only adults.
|
Consistent with previous research on the overall population of wireless-only adults (e.g., Blumberg and Luke 2007
), wireless-only young adults, wireless-only low-income adults, and wireless-only low-income young adults were more likely than their counterparts with landlines to be renting their homes, living alone or with unrelated adult roommates, and living in metropolitan areas (see table 2). Within all the three subgroups, wireless-only adults were more likely to be male, to have completed at least high school, to be working at a job or business, and to be living in the South when compared with adults with landlines.
|
Wireless-only young adults were more likely to be low-income than were young adults with landlines. Wireless-only low-income adults were more likely to be young than were low-income adults with landlines. Wireless-only young adults were more likely to be non-Hispanic white and less likely to be Hispanic than were young adults with landlines.
|
Coverage Bias in Landline Surveys of Young Adults |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
Relative to young adults with landlines, wireless-only young adults were more likely to binge drink, smoke, and engage in regular physical activity (see table 3 and the unadjusted bias analysis in table 4). They were more likely to have had an unmet need for medical care due to cost and less likely to have a usual place for medical care, though no significant differences were observed for health insurance status. They were also less likely to be obese. No other significant differences in health status were noted between wireless-only young adults and young adults with landlines.
|
|
For most of these health-related behaviors and health care access measures, the potential coverage bias due to exclusion of wireless-only and phoneless young adults was statistically significant even after adjustment for demographic differences. Among the 13 measures examined, the potential bias would be the greatest for alcohol-related surveys of young adults. Such surveys would be expected to underestimate the prevalence of binge drinking by 3.2 percentage points despite demographic adjustments.
The potential coverage bias was greater than 2 percentage points for HIV testing. Wireless-only young adults were more likely to have ever been tested for HIV, and telephone surveys that only include landlines would be expected to underestimate the prevalence of HIV testing among young adults by 2.1 percentage points after adjustment for demographic differences.
For young adults, no other measures of potential coverage bias exceeded 2 percentage points, which is a threshold described in the past as an indicator of "practical significance" (Blumberg, Luke, and Cynamon 2006).
|
Coverage Bias in Landline Surveys of Low-Income Adults |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
Similar to wireless-only young adults, wireless-only low-income adults were more likely to binge drink and smoke than were their counterparts with landlines (see tables 3 and 4). However, wireless-only low-income adults had generally better-reported health than low-income adults with landlines. They were more likely to report excellent or very good health, more likely to exercise, less likely to be obese, and less likely to have been diagnosed with diabetes. Relative to low-income adults with landlines, wireless-only low-income adults were more likely to be uninsured, more likely to have experienced financial barriers to needed medical care, less likely to have a usual place for care, and less likely to have received a flu vaccination during the past year. They also were more likely to have been tested for HIV.
Most of these differences result in the potential for statistically significant coverage bias in landline surveys, even when adjusted for demographic differences. However, none of the measures of potential coverage bias exceeded 2 percentage points once adjustments were made. The greatest potential coverage bias was found for flu vaccination (overestimated by 1.7 percentage points), smoking (underestimated by 1.4 percentage points), and obesity (overestimated by 1.3 percentage points).
|
Coverage Bias in Landline Surveys of Low-Income Young Adults |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
Analyses limited to both low-income and young adults reveal that wireless-only adults in this group differ from those with landlines on only four health measures (see table 3). Wireless-only low-income young adults were more likely to binge drink, exercise, and have financial barriers to care. They were also less likely to be obese.
With adjustments for demographic differences, landline telephone surveys of low-income young adults are expected to underestimate binge drinking and regular exercise and to overestimate obesity and the receipt of influenza vaccinations. Significant bias would also be expected for estimates of current smoking despite no statistically significant difference in smoking behavior between wireless-only low-income young adults and their counterparts with landlines. The potential for coverage bias in smoking estimates occurs because nearly half (44.8 percent) of the low-income young adults without any type of telephone service were smokers.
For these five health indicators, the potential magnitude of the coverage bias was greater than or equal to 2 percentage points. The potential for practically significant coverage bias extends also to measures of HIV testing among low-income young adults, though this estimate of bias was only marginally significant (p =.06) in part due to larger estimates of variability resulting from small available sample sizes.
|
Implications |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
Wireless-only adults differ demographically from adults with landlines, even when limiting one's focus to young adults, low-income adults, or low-income young adults. These differences are similar to the differences noted in other studies that have focused broadly on all adults (e.g., Blumberg and Luke 2007
; Tucker, Brick, and Meekins 2007), which is perhaps not surprising given that the three subgroups examined here make up 24–51 percent of the total population of wireless-only adults. Assuming that proper sample weighting accounts for these known differences, coverage bias exists only when: (1) there are unknown or unmeasured characteristics that differ between adults with and without landlines, and (2) these unknown or unmeasured characteristics are related to the substantive measures of interest. This report is one of the first to demonstrate that the increase in the prevalence of wireless-only adults will lead to nonnegligible coverage biases in landline telephone surveys even after adjusting for demographic differences.2 These results suggest that there are unknown differences between adults with and without landlines and that these unknown differences are related to lifestyle preferences, such as binge drinking or smoking. Perhaps adults without landlines differ from adults with landlines in extroversion, in the size of their social networks, or in their participation in social activities. These characteristics are predictive of willingness to participate in cell-phone surveys (Vehovar and Callegaro 2007) and of binge drinking and smoking (Wechsler et al. 1995; Rigotti, Lee, and Wechsler 2000; Munafo, Zetteler, and Clark 2007), and they may also be predictive of decisions to substitute wireless telephones for landlines or to live without telephones altogether. To our knowledge, however, no nationally representative surveys produce estimates for such characteristics, limiting their utility in sample weighting.
The present findings highlight the need to carefully develop sample weights for landline telephone surveys, using multiple demographic control totals. Limiting the analysis to low-income young adults did not reduce the magnitude of the potential coverage bias, suggesting that coverage bias for health indicators is not simply the result of the younger age and lower income of adults without landlines. Adjustments that included the eight other demographic characteristics, however, did attenuate the magnitude of the potential biases. For many indicators, in fact, the bias was reduced sufficiently that any practical or programmatic implications of the bias would be minimal. For example, with proper weighting and sufficient demographic controls, landline telephone surveys of health conditions (e.g., asthma, diabetes, psychological distress) and health insurance may still be able to ignore issues related to coverage bias, even when focusing on young adults or low-income adults.
However, we urge caution when conducting (or interpreting results from) landline telephone surveys of health risk behaviors and HIV testing, and perhaps for surveys about exercise and obesity as well. The present findings of significant coverage bias for these lifestyle preferences among young adults are consistent with other research. For example, a study by the Pew Research Center revealed that national estimates from a landline survey of young adults (18–25 years) produced biased estimates of alcohol consumption and exercise during the previous week, relative to blended estimates from a landline and cell-phone survey (Keeter et al. 2007). Estimates for smoking, however, were unbiased. Another cell-phone survey conducted in three states as a pilot project for the Behavioral Risk Factor Surveillance System revealed that wireless-only adults were more likely to smoke, were more likely to have been tested for HIV, and were less likely to have health insurance than were adults with landlines (Link et al. 2007). Similar caution may also be warranted when conducting (or interpreting results from) landline telephone surveys of other lifestyle preferences, such as illicit drug use, nonhealth-related risk behaviors, or participation in social activities.
The inclusion of wireless telephone numbers in telephone sampling frames has been suggested as a solution to the problem of coverage bias (e.g., Brick et al. 2007). The present findings reveal that at least 86 percent of young adults without landlines have wireless telephones in their homes and may be accessible through cell-phone surveying. The vast majority of low-income adults without landlines and low-income young adults without landlines also have access to wireless telephones (77.2 percent and 84.8 percent, respectively). Surveys conducted solely on wireless telephones may even be a cost-effective way to survey young adults, as more than 68.5 percent of young adults have access to a wireless telephone.
In closing, however, it is worth noting that coverage bias is not the only source of potential bias in telephone surveys. Nonresponse bias and systematic error due to mode effects may be even greater for cell phone surveys than for landline surveys, and survey researchers may therefore return to landline telephone surveys with a hope that coverage bias can be further attenuated by statistical adjustments. The 2007 NHIS includes several new questions on the frequency of landline and wireless telephone use (for households with both types of service) and on the possible transient nature of wireless-only status (cf. Keeter 1995). Perhaps future analyses of data from these new questions will suggest worthwhile statistical techniques to account for the coverage biases highlighted here.
|
Footnotes |
|---|
STEPHEN J. BLUMBERG AND JULIAN V. LUKE are with the National Center for Health Statistics, Centers for Disease Control and Prevention, 3311 Toledo Road, Hyattsville, MD 20782, USA. The findings and conclusions are those of the authors and do not necessarily represent the views of the National Center for Health Statistics, Centers for Disease Control and Prevention.
1 In the NHIS, a family roster is obtained, and for most families, each adult in the family has an equal probability of being selected by the computer-assisted personal interviewing (CAPI) system's random selection process. There is one exception: Any Black, Asian, or Hispanic adult aged 65+ years is given twice the chance of being selected as the sample adult as any other adult in the family. 
2 To our knowledge, the only other published studies demonstrating significant and nonnegligible coverage biases in landline telephone surveys after adjusting for demographic differences are Pew Research Center studies on technology use and technological sophistication (e.g., Pew Research Center 2007).
|
References |
|---|
|
TOPAbstractIntroductionData SourceAnalysis ProceduresPrevalence and Demographic...Coverage Bias in Landline...Coverage Bias in Landline...Coverage Bias in Landline...ImplicationsReferences |
|---|
Blumberg Stephen J., Luke Julian V. Wireless Substitution: Early Release of Estimates Based on Data from the National Health Interview Survey, July–December 2006 (2007) Hyattsville, MD: National Center for Health Statistics. Available online at http://www.cdc.gov/nchs/nhis.htm.
Blumberg Stephen J., Luke Julian V., Cynamon Marcie L. "Telephone Coverage and Health Survey Estimates: Evaluating Concern about Wireless Substitution." American Journal of Public Health (2006) 96:926–931.
Blumberg Stephen J., Luke Julian V., Cynamon Marcie L., Frankel Martin R., Forthcoming. "Recent Trends in Household Telephone Coverage in the United States." In: Advances in Telephone Survey Methodology—Lepkowski, et al, eds. New York: J. W. Wiley and Sons, Inc.
Botman Steven L., Moore Thomas F., Moriarity Christopher L., Parsons Van L. "Design and Estimation for the National Health Interview Survey, 1995–2004." In: Vital and Health Statistics 2 (2000) 130:1–31.
Brick J. Michael, Brick Pat D., Dipko Sarah, Presser Stanley, Tucker Clyde, Yuan Yangyang. "Cell Phone Survey Feasibility in the U.S.: Sampling and Calling Cell Numbers versus Landline Numbers." Public Opinion Quarterly (2007) 71:23–29.
Graubard Barry I., Korn Edward L. "Predictive Margins with Survey Data." Biometrics (1999) 55:652–659.[CrossRef][ISI][Medline]
Keeter Scott. "Estimating Telephone Noncoverage Bias from a Phone Survey." Public Opinion Quarterly (1995) 59:196–217.[Abstract]
Keeter Scott, Kennedy Courtney, Tompson Trevor, Mokrzycki Mike, Clark April. What's Missing from National RDD Surveys? The Impact of the Growing Cell-Only Population. (2007) Paper presented at the annual meeting of the American Association for Public Opinion Research, Anaheim, CA.
Link Michael W., Battaglia Michael P., Frankel Martin R., Osborn Larry, Mokdad Ali H. Conducting Public Health Surveys over Cell Phones: The Behavioral Risk Factor Surveillance System Experience. (2007) Paper presented at the annual meeting of the American Association for Public Opinion Research, Anaheim, CA.
Munafo Marcus R., Zetteler Jessica I., Clark Taane G. "Personality and Smoking Status: A Meta-Analysis." Nicotine and Tobacco Research (2007) 9:405–413.[CrossRef]
Pew Research Center for the People and the Press. "A Portrait of ‘Generation Next'." (2007) Available online at: http://people-press.org/reports/pdf/300.pdf.
Rigotti Nancy A., Lee Jae Eun, Wechsler Henry. "US College Student’ Use of Tobacco Products: Results of a National Survey." Journal of the American Medical Association (2000) 284:699–705.
Schiller Jeannine S., Martinez Michael, Barnes Patricia. Early Release of Selected Estimates Based on Data from the 2005 National Health Interview Survey (2006) Hyattsville, MD: National Center for Health Statistics. Available online at: http://www.cdc.gov/nchs/about/major/nhis/released200606.htm.
Tucker Clyde, Brick J. Michael, Meekins Brian. "Household Telephone Service and Usage Patterns in the United States in 2004: Implications for Telephone Samples." Public Opinion Quarterly (2007) 71:3–22.
US Census Bureau. "Poverty Thresholds 2006." (2007) Available online at http://www.census.gov/hhes/www/poverty/threshld/thresh06.html.
Vehovar Vasja, Callegaro Mario. Social Aspects of Mobile Phone Usage and Their Impact on Survey Cooperation. (2007) Paper presented at the annual meeting of the American Association for Public Opinion Research, Anaheim, CA.
Wechsler Henry, Dowdall George W., Davenport Andrea, Castillo Sonia. "Correlates of College Student Binge Drinking." American Journal of Public Health (1995) 85:921–926.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||