Has the National Do Not Call Registry Helped or Hurt State-Level Response Rates?
A Time Series Analysis
MICHAEL W. LINK is a Senior Survey Methodologist and ALI H. MOKDAD is a Branch Chief in the National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention. DALE KULP is President and CEO of Marketing Systems Group. ASHLEY HYON is an Account Executive with Marketing Systems Group.
Address correspondence to Michael W. Link; e-mail: MLink{at}cdc.gov.
 |
Abstract |
|---|
 TOP Abstract Background on the National...Data and MethodsResultsDiscussionReferences |
|---|
By the end of the initial registration period on August 31, 2003, the National Do Not Call Registry (DNC Registry) had registered more than 50 million telephone numbers. Approximately 18 months later that number had increased to more than 91 million. The impact of the DNC Registry on survey response rates, however, is largely unknown. Some researchers speculate that the registry could make it easier to distinguish between telephone survey interviewers and telemarketers. Other researchers argue that a significant portion of DNC registrants may not make such distinctions and would prefer instead to reduce all unsolicited calls from marketers and interviewers alike. Case outcomes from nearly 4.5 million telephone numbers called between January 1, 2002, and June 30, 2005, as part of the Behavioral Risk Factor Surveillance System were analyzed. Using trend analyses and autoregressive integrated moving average (ARIMA) time series modeling, we assessed the impact of the DNC Registry on state-level monthly response rates in 47 states. Our findings indicate that once pre-DNC Registry trends in response rates and other potential covariates are accounted for, the national Do Not Call rules have had no significant impact on state-level response rates in either a positive or negative direction.
On October 1, 2003, the Federal Trade Commission (FTC) launched the National Do Not Call Registry (DNC Registry). Registration is ongoing, although by the end of the initial registration period the DNC Registry contained more than 50 million telephone numbers.1 And just 18 months later, that number had increased to more than 91 million. The registry was developed "to offer consumers a choice regarding telemarketing calls," allowing people to register their landline and cellular telephone numbers to reduce unsolicited calls from certain segments of the telemarketing industry (Federal Trade Commission 2003a
). Calls from survey researchers are not covered by the DNC Registry rules; however, because such a large segment of the U.S. population registered their telephone numbers, the survey industry may still experience consequences of some type. On the one hand, the registry may "clear the waters" of telephone solicitation, including calls related to sugging ("selling under the guise of research") and frugging ("fundraising under the guise of research"), thereby making it easier for legitimate survey researchers to distinguish themselves from telemarketers (Schulman 2005). On the other hand, a significant portion of DNC registrants may not make such distinctions and would prefer instead to reduce all unsolicited calls from marketers and interviewers alike. The former scenario could help improve the likelihood of obtaining telephone survey participation, whereas the latter scenario would not affect the already declining response rates in telephone surveys. Our evaluation assesses the initial impact of the DNC Registry on one of the largest ongoing random digit dial (RDD) telephone surveys, the Behavioral Risk Factor Surveillance System (BRFSS). This state-based system, which is run with assistance from the Centers for Disease Control and Prevention (CDC), measures behavioral risk factors in the noninstitutionalized population of adults aged 18 years or older. The survey is conducted monthly in all 50 states, the District of Columbia, Puerto Rico, Guam, and the Virgin Islands. The BRFSS collects state-specific data on preventive health practices and risk behaviors linked to chronic diseases, injuries, and preventable infectious diseases in the adult population. In the present study we analyzed outcomes from nearly 4.5 million sampled telephone numbers called over a 42-month time frame (from January 1, 2002, to June 2005) and assessed the initial impact of the DNC on monthly state-level telephone survey response rates in 47 states.
|
Background on the National DNC Registry |
|---|
|
TOPAbstractBackground on the National...Data and MethodsResultsDiscussionReferences |
|---|
The DNC Registry was designed to give people who own telephones a mechanism for reducing unsolicited calls from certain segments of the telemarketing industry. A telemarketer is defined by the FTC as "any person who, in connection with telemarketing, initiates or receives telephone calls to or from a customer" (Federal Trade Commission 2003b
). Individuals can register up to three personal (not business) telephone numbers through either a Web site (which requires an active e-mail address to complete registration) or by calling a toll-free number. Telephone numbers remain on the registry for five years from the date registered. Enforcement of the registry’s policies falls to the FTC, the Federal Communications Commission (FCC), and the individual states. The DNC Registry does not apply to all unsolicited telephone calls. In particular, calls from or on behalf of charities, political organizations, or survey researchers are not covered under the registry’s rules, nor are companies with which the individual has an existing business relationship or has expressed in writing interest in receiving their calls. However, calls from companies selling goods or services under the guise of conducting a survey are required to abide by the registry’s rules.
Because of the relative newness of the national Do Not Call list, little is known about the types of people who register and how they might differ from people who choose not to register. DNC registrants are behaviorally different than nonregistrants simply by virtue of having registered their telephone numbers. At a minimum, this specific action indicates that these individuals want to limit unsolicited telephone calls from telemarketers, but they may also want to limit unsolicited calls in general (including those from survey researchers). Conversely, a subset of DNC registrants may want to limit unsolicited calls from telemarketers but be willing to entertain calls from other organizations, including legitimate survey organizations. These registrants may actually be more willing to participate in survey efforts if other unsolicited calls are reduced or eliminated. This is the viewpoint shared by the American Association for Public Research (AAPOR), which issued several statements and an amicus curiae brief in support of the DNC Registry (Schulman 2005).2 Another possibility is that both sets of motivations may be at work.
|
Data and Methods |
|---|
|
TOPAbstractBackground on the National...Data and MethodsResultsDiscussionReferences |
|---|
To assess the potential impact of the DNC Registry on survey response rates, we calculated monthly state-level response rates by aggregating information from the call records of 4,480,469 telephone numbers called as a part of the BRFSS between January 1, 2002, and June 30, 2005. This time frame includes the 21 months preceding and the 21 months following the October 1, 2003, start date of the DNC Registry. The data came from 47 U.S. states.3 As a state-based survey with considerable variability across states in terms of response rates and percentage of households registered on the DNC Registry, the BRFSS is uniquely suited to provide a robust analysis of the potential impact of the registry on survey participation. Key data collection protocols are standardized across these states, which helps to facilitate cross-state comparisons. For instance, all unanswered telephone numbers and contacted nonrefusing households received a minimum of 15 callbacks on different days of the week and at different times of the day (or in accordance with a scheduled appointment). Attempts were made to convert people who initially refused to participate with the exception of hostile sample members. Finally, a common set of interim and final disposition codes was used across states to track the disposition of a telephone number on each call attempt.
MEASURES USED IN ANALYSIS
Monthly response rates were calculated using AAPOR response rate 4 (AAPOR 2004). These response rates reflect the proportion of all confirmed and potentially eligible sample members for whom an interview has been completed. To minimize the potential impact of variation in sample designs across these monthly periods, only cases from 1+ residential listed blocks screened by Marketing Systems Group as being potential households were included in the analysis. The data were also adjusted so that each monthly sample reflected the mean proportion of listed residential numbers and unlisted (1+ block) numbers across the 2002–2005 time frame.
We calculated the percentage of households in each state listed on the DNC Registry on a quarterly basis from October 2003 through June 2005.4 These estimates were calculated by dividing the number of DNC telephone numbers registered in a state (adjusted to account for the fact that households can register more than one number) by the number of households with telephones in the state. The number of telephone numbers in each state registered on the DNC Registry came from summary statistics provided by Marketing Systems Group, which operated under FCC guidelines for access to the DNC Registry database. The number in each state was then adjusted to account for the fact that households can register more than one telephone number per household. A national survey found that DNC registrants listed on average 1.3 telephone numbers per household (Quarles and Schnurr 2004). We divided the number of DNC-listed telephone numbers by this factor to convert the number of DNC registered telephone numbers into an estimate of the number of households with one or more registered telephones.5 Using state-based estimates for the number of households with telephones developed from the Current Population Survey, Social and Economic Supplement 2003 (U.S. Census Bureau 2004), we developed the quarterly estimates of the percentage of households in each state with at least one telephone number listed on the DNC Registry. These quarterly estimates were applied to each month within the respective quarter. To account for the possibility that people who registered between June and September 2003 might have thought the rules went into effect when they registered, we imputed the percentage of registrants in these four months by simply assuming a linear increase from no registrants in May to the empirical percentage calculated for October 2003.
Three other potential covariates were calculated and used as controls in the final models:
Level of Effort.
Fluctuations in response rates over time may be attributable to variations in the level of effort put into collecting these data from month to month. We initially calculated the monthly level of effort as the total number of dialings made across all cases during a month divided by the total number of sampled cases that received at least one telephone call. Both the number of dialings made during a month and the number of telephone numbers released for calling during a month can affect response rates. For example, all other factors being equal, we would expect higher response rates in situations where a smaller number of cases received a relatively large number of calls, compared with a larger sample with fewer calls per case. Since response rates tend to level off as the number of calls per case increases, the final variable used in analysis was calculated as the square root of the dialings per case measure.
Use of Advance Letters.
Prenotification letters have been shown in a number of studies to improve survey response rates (Goldstein and Jennings 2002; Hembroff et al. 2005; Link and Mokdad 2005). States are not required to use advance letters as a part of the BRFSS; however, during the 2002–2005 time frame, seven states began using advance letters as a part of their state-based protocol. A dichotomous indicator was developed to identify months when advance letters were and were not used in these states.
Change in Data Collector.
Because the BRFSS is a state-run survey, data collection is not centralized with a single contractor. States are free to collect the data using their own facilities or to contract the work to a data collection vendor. When a state changes vendors there may be an effect (either positive or negative) on response rate trends, depending on the new contractor’s ability to complete interviews. During the 2002–2005 time period, six states changed data collection contractors. A dichotomous variable was developed to indicate the months when the new contractor was used.
STATISTICAL ANALYSES
The analyses were conducted in three parts. First, we assessed the extent to which state-level participation in the DNC Registry changed over the 42 months examined here. Second, ordinary least square (OLS) regression techniques were used to describe significant trends in monthly state-level response rates and level of effort. Third, autoregressive integrated moving average (ARIMA) models were developed to examine the impact of the DNC Registry and other potential covariates on monthly trends in state-level response rates. All analyses were conducted using SPSS 13.0 (SPSS Inc., Chicago, IL).
|
Results |
|---|
|
TOPAbstractBackground on the National...Data and MethodsResultsDiscussionReferences |
|---|
ESTIMATED PERCENTAGE OF HOUSEHOLDS ON DNC REGISTRY
The percentage of households with at least one telephone number on the National DNC Registry grew substantially during the first 21 months of the list’s operation (see figures 1 and 2). As of October 2003, the median percentage across states was 34 percent. By June 2005 the state median percentage had grown to 63 percent, ranging from 87 percent in Colorado to 39 percent in Indiana, with considerable variability noted across states. The mean percentage of registrants was somewhat higher in northeastern states (67 percent) compared with states in the South (58 percent), Midwest (62 percent), and West (62 percent); however, these regional differences were not statistically significant (F = 1.11, df = 3, p = .351).
|
|
TRENDS IN STATE-LEVEL RESPONSE RATES
Table 1 provides the OLS regression coefficients for state-level trends in monthly response rates for the period January 2002 through June 2005. States varied considerably in the level of change in their monthly response rates over time. Of the 47 states examined, 33 (70 percent) showed significant declining trends in response rates, ranging from a 4.3 percent (Alabama) decline per year to less than a 1 percent decline per year (Mississippi and Pennsylvania) when the regression coefficient is multiplied by 12 months. The median value among states with a significant decline in response rates was –2.2 percent over the course of a year. Four states showed a significant (p < .05) increase in response rates over time: Alaska (+1.8 percent), South Carolina (+3.7 percent), Minnesota (+4.1 percent), and South Dakota (+5.4 percent). A fifth, New Mexico, showed positive trending at the p < .10 level. The remaining states showed no significant trends in monthly response rates over this 42-month period.
|
Figure 3 shows changes over time in the monthly state median response rate and the percentage of households on the DNC Registry. The median percentage of households on the DNC Registry increased sharply from its inception in 2003, while median response rates continued to decline during this same period. However, there does not appear to have been an upward or downward spike in median response rates at the time the DNC Registry was launched, nor as the percentage of households on the DNC Registry increased.
|
A large percentage of states also showed significant trends in level of effort over this 42-month period. The ratio of the number of calls to number of total cases sampled (before taking the square root) increased significantly in 38 of 47 (80 percent) of the states (data not shown). Among those with a significant trend, the yearly increase in the number of calls per case ranged from 0.5 calls (Nevada) to 3.1 calls (Wisconsin) with a state-level median increase of 1.4 calls. The nine states that did not show a significant trend in calls per case were Connecticut, Florida, Maryland, Massachusetts, New Hampshire, New Jersey, Ohio, Rhode Island, and Vermont. None of the states showed a significant decline in level of effort over this period.
ARIMA MODELING OF RESPONSE RATE TRENDS
Looking more closely at response rate trends and the factors that might affect them, we used time series methods to determine if changes in response rates were due to increasing registration on the DNC Registry, fluctuations in the monthly level of effort, use of advance letters, changes in data collectors, or simply a continuation of prior trends caused by other factors. Given the fact that nearly three-quarters of the states showed some type of significant trend in their response rates and the requirement for ARIMA modeling that these series be stationary (i.e., have a constant mean and variance over time), the response rate values were differenced. This was accomplished by subtracting the previous month’s response rate value from the current value. In order to allow comparisons across states, differencing was used for all of the states, even those where the series was already stationary. We then identified the appropriate ARIMA model by examining the values of the autocorrelation and partial autocorrelation functions of each state’s series. First-order autoregressive models (1,1,0) were identified for 29 of the states (62 percent), and a simple differenced model (0,1,0) was identified for the remaining 18 states. Each model was estimated using the differenced response rate values as the dependent variable, and the DNC Registry, level of effort, advance letter, change in data collector, and first-order autoregressive term (where appropriate) as independent variables.
Results of the ARIMA models are shown in table 1. The findings indicate that changes in the percentage of households on the National Do Not Call Registry appear to have had little impact on changes in differenced response rates over time. For 23 states (49 percent), the direction of the DNC Registry coefficient was positive, while for the remaining 24 states the coefficient was negative. None of the states, however, showed a statistically significant relationship (p < .05) between changes in the state-level percentage of households on the DNC Registry and the differenced monthly response rates. For two states, the data are suggestive of a positive relationship, when the significance levels are relaxed to p < .10. In California and Florida, an 8 to 9 percentage point increase in the percentage of households on the DNC Registry resulted in an increase in response rates of approximately 1 percentage point.
All three of the other covariates were significantly related to response rates in some, but not all, of the states. Level of effort showed a statistically significant positive relationship with response rates in 9 of the 47 states (19 percent) and a significant negative relationship in two states (4 percent). Increased level of effort had the largest effect in Minnesota. From January 2002 to June 2005 the average number of calls per case increased from 6.1 to 11.7. This resulted in an increase in the differenced response rates of 14.6 percentage points.
Likewise, use of advance letters had significant effects in some but not all states where prenotification was used. Of the seven states that started using advance letters between January 2002 and June 2005, five showed a significant positive effect on changes in monthly response rates. Among the four states where advance letters were used but there was no change in the data collector, the impact of the letters on monthly changes in response rates ranged from +5.1 percentage points (Maine) to +9.5 percentage points (Oregon). In South Carolina, where the use of advance letters coincided with a change in data collector, the increase in response rates was +18.8 percentage points from these two factors combined. Use of advance letters did not appear to affect monthly response rates significantly in Connecticut or North Carolina. None of the six states that changed data collectors showed a significant impact of this change at the p < .05 level. The results from Texas are suggestive of a decrease in response rates of just over 6 percentage points when the significance level is relaxed to p < .10.
|
Discussion |
|---|
|
TOPAbstractBackground on the National...Data and MethodsResultsDiscussionReferences |
|---|
Like many other RDD surveys, response rates for the BRFSS have declined significantly in a majority of states. While a few states have managed to counter these trends or at least maintained fairly steady rates of participation, most have experienced declines of approximately 2 percentage points per year since January 2002. This rate of decline is similar to that seen in the University of Michigan’s Survey of Consumer Attitudes, which reported response rate declines averaging 1.5 percentage points between 1996 and 2003 (Curtin, Presser, and Singer 2005
). Identifying the factors related to the decline in survey participation is critical to the development of potential strategies for stemming or reducing the rate of this decline. The National DNC Registry is one of those potential factors. The DNC Registry is a relatively rare phenomenon in survey research in that it has the potential to affect the entire RDD sampling frame and to do so very rapidly. For example, at the time the DNC Registry was initiated, it contained more than 50 million telephone numbers. Within 18 months, that number nearly doubled. Although these effects might not have a legal basis, as survey research calls are exempt from DNC calling restrictions, there could still be more immediate effects on the motivation of sample members to respond to a survey inquiry. In contrast, most other factors that have affected survey research efforts have tended to be introduced more gradually. A case in point is the use of call-screening technology, such as telephone answering machines and Caller-ID, which took a number of years to be adopted by a substantial portion of the public (Link and Oldendick 1999; Oldendick and Link 1994; Tuckel and O’Neill 2002). Because registration on the DNC Registry is ongoing and is apparently still increasing in many states, it is unclear at what point registration will level off. Based on the findings of this study, the DNC Registry appears to have had no significant (p < .05) independent effect (positive or negative) in any of the 47 states examined. The findings from California and Florida are suggestive of a positive relationship if significance is evaluated at the p < .10 level. Though a majority of states have experienced significant declines in their response rates during the 21-month periods before and after the official launching of the DNC Registry, and a few have shown trends in a positive direction, neither the positive nor the negative trends appear to be related to the launch or growth of the DNC Registry.
Changes in response rates during this period were more likely to be related to other covariates, but the effects of these factors were not uniform across all states. For instance, while level of effort has increased in a majority of states, effort has had a significant positive impact on response rates in only nine of them. In two states (Minnesota and South Dakota) the significant increase in level of effort over time has resulted in a significant positive trend in response rates, and in three others (Ohio, New Mexico, and Tennessee) response rates have remained stable. For these states, increases in effort over time appear to have either improved response rates or at a minimum held them reasonably constant. In contrast, in four states (Delaware, Maine, North Carolina, and Wyoming) response rates have shown a significant downward trend, despite the positive relationship between level of effort and response rates in the ARIMA model (data not shown) and a positive relationship between level of effort and response rates. In two states (Arizona and North Dakota) the relationship between level of effort and response rates is negative. For these last six states, improvement in the number of call attempts over time does not appear to have been enough to reverse the downward trend in survey participation. It may be that the additional calls were not being placed during the most productive call-scheduling windows, when respondents were most likely to be at home. Unfortunately, the data analyzed here do not contain information about the days of the week and times of the day in which calls were made, which would allow us to explore this hypothesis further.
We considered two other covariates, advance letters and changes in data collection agencies. The positive effects of advance letters and the magnitude of those effects are similar to those reported by other research in this area (Goldstein and Jennings 2002; Hembroff et al. 2005; Link and Mokdad 2005). In contrast, the effect of changing data collection organizations appears to have had very little effect on response rates.6
This study had several limitations. First, we measured changes in survey participation at an aggregate (state) level, rather than at the individual level. Unfortunately, the regulations currently governing access to DNC Registry information do not permit the registry to be used to identify individual registrants. Second, the steady decline in response rates in most states, despite increased level of effort, suggests that other potential operational and societal factors beyond those controlled for here are affecting these rates. Third, the time frame examined was of modest duration (42 months). Unfortunately, including additional months prior to January 2002 is complicated by the fact that in 2002 the disposition codes for the BRFSS were expanded from 15 to more than 30 to align them more closely with the categories recommended by AAPOR. While this improved the accuracy of BRFSS response rates, it also made comparisons with earlier time periods more problematic. Fourth, we considered only the potential impact of the National DNC Registry and did not take into account the potential influence of do not call lists that states might maintain separately from the national list. Unlike the national registry, there is no central historical archive of these data and in some instances no accessible historical archive at the state level either, particularly once states began sharing their lists with the national registry. As of the end of March 2005, only three states maintained state-based lists that were not shared with the National DNC Registry.
Because of the scope of the DNC Registry and the growing number of telephone lines being registered, it is imperative that research in this area continue. Not only do survey researchers need to understand the short- and long-term impact of the DNC Registry on survey participation rates, but we also need to develop a much better understanding of the registrants themselves. How do registrants differ from nonregistrants in their attitudes, opinions, and behaviors? How might these differences affect the estimates we derive from telephone surveys? Unfortunately, without a means of identifying the universe of DNC registrants, research efforts will be limited to self-reports, which makes research by telephone exceedingly difficult. The problem is similar to that faced by researchers trying to understand the use and impact of call screening with Caller-ID. The universe of users is unknown, and telephone-based studies face the problem of collecting information only from the people who are willing to participate in telephone surveys. An alternative mode for collecting data, such as face-to-face surveys, may be the only reliable way to collect such information in the foreseeable future.
Overall, the launching of the National DNC Registry on October 1, 2003, has not helped survey research efforts in the short term, nor does it appear to have hurt them. Like other factors affecting survey research, however, it may take a longer period for the registry to have an effect. If the number of unwanted calls to households diminishes, the public may be more willing to participate in telephone surveys. It is important, therefore, that the potential impact of the DNC Registry continue to be monitored in order to determine whether the registry is capable of slowing or reversing downward trends in response, or whether continued registration only makes a bad situation worse.
|
Footnotes |
|---|
The findings and conclusions in this article are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention or Marketing Systems Group.
1. The initial registration period for the DNC Registry ran from June 1 through August 31, 2003. Telephone numbers registered during this period were the first to be covered under the registry’s rules, which went into effect for covered organizations on October 1, 2003 (Federal Trade Commission 2003a). 
2. The AAPOR brief of amicus curiae was filed in the case of Mainstream Marketing Services, Inc. et al. versus Federal Trade Commission et al. (case #03-1429).
3. Hawaii, Illinois, and the District of Columbia were excluded because of significant gaps in monthly data collection during the time period examined. Michigan was excluded because the state uses a quarterly rather than a monthly sampling design. The DNC Registry does not apply to, or participation was less than 0.5% in, Puerto Rico, the Virgin Islands, and Guam, which also administer the BRFSS.
4. Telemarketers and list vendors covered under these rules were initially required to obtain updated information from the DNC Registry every three months; hence, quarterly measures were calculated here. Beginning in January 2005, however, covered organizations were required to obtain updated list information every 31 days (Federal Trade Commission 2003a).
5. One limitation of this approach is that the 1.3 figure is a national estimate, which most likely varies across individual states. Unfortunately, no state-level measures are currently available. State-by-state variation, however, is expected to be small and similar to the variability in reports of the number of residential telephone lines in a household, which averaged 1.8 to 2.2 residential lines per household in the states examined here.
6. The data collection agency most likely has a larger effect on response rates, but data collectors are perfectly confounded with states, and their effect can therefore not be analyzed here.
|
References |
|---|
|
TOPAbstractBackground on the National...Data and MethodsResultsDiscussionReferences |
|---|
American Association for Public Opinion Research (AAPOR). (2004) Standard Definitions: Final Dispositions of Case Codes and Outcome Rates for Surveys 3d ed (AAPOR, Lenexa, KS).
Curtin Richard, Presser Stanley, Singer Eleanor. (2005) Changes in Telephone Survey Nonresponse over the Past Quarter Century. Public Opinion Quarterly 69:87–98.
Federal Trade Commission. (2003a) Do Not Call—FAQs. Available online at http://www.ftc.gov/bcp/conline/pubs/alerts/dncalrt.htm (accessed June 1, 2005).
———. (2003b) Entity Definitions. Available online at https://telemarketing.donotcall.gov/Entity.aspx (accessed June 1, 2005).
Groves Robert M. and Couper Mick P. (1998) Nonresponse in Household Interview Surveys(Wiley, New York).
Goldstein Kenneth M. and Kent Jennings M. (2002) The Effect of Advance Letters on Cooperation in a List Sample Telephone Survey. Public Opinion Quarterly 66:608–17.[CrossRef][Web of Science]
Hembroff Larry A., Rusz Debra, Rafferty Ann, McGee Harry, Ehrlich Nathaniel. (2005) The Cost-Effectiveness of Alternative Advance Mailings in a Telephone Survey. Public Opinion Quarterly 69:232–45.
Link Michael W. and Mokdad Ali H. (2005) Advance Letters as a Means of Improving Respondent Cooperation in Random Digit Dial Studies: A Multistate Experiment. Public Opinion Quarterly 69:572–87.
Link Michael W. and Oldendick Robert W. (1999) Call Screening: Is It Really a Problem for Survey Research? Public Opinion Quarterly 63:575–89.
Oldendick Robert W. and Link Michael W. (1994) The Answering Machine Generation. Public Opinion Quarterly 58:264–73.
Quarles Rebecca C. and Schnurr Robert W. (2004) How the Do-Not-Call Registry Might Affect Survey Research: A Pilot Study. Paper presented at the annual meeting of the American Association for Public Opinion Research, Phoenix, AZ.
Schulman Mark A. (2005) Appeals Court Overturns Major ‘Do-Not-Call’ Threat. AAPOR News 32:1–6 Available online at http://www.aapor.org/default.asp?page=news_and_issues/aapor_news (Spring): (accessed October 31, 2005).
Tuckel Peter and O’Neill Harry. (2002) The Vanishing Respondent in Telephone Surveys. Paper presented at the Council for Marketing Opinion Research Respondent Cooperation Workshop, New York.
U.S. Census Bureau. (2004) Current Population Survey, Annual Social and Economic Supplement, 2003 Available online at http://www.bls.census.gov/cps/cpsmain.htm (accessed June 1, 2005).
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  C. D. Delnevo, D. A. Gundersen, and B. T. Hagman Delnevo et al. Respond to "Topical Threats to Epidemiology" Am. J. Epidemiol., January 1, 2008; 167(1): 23 - 24. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||