The leading association
of public opinion and
survey research professionals
American Association for Public Opinion Research

Nonresponse in RDD Cell Phone Surveys

Previous: Coverage and
for RDD Cell Phone Surveys

Table of Contents

Next: Measurement in 
  RDD Cell Phone Surveys

One of the most problematic features of general population RDD surveys in the U.S. is their low response rates. This problem applies to both landline RDD surveys and cell phone RDD surveys.

As measured in early comparison studies, cell phone survey response rates were approximately 10 percentage points less than response rates in comparable RDD landline surveys (Steeh and Piekarski, 2008). However, evidence from more recent dual frame surveys suggests the difference has narrowed to approximately five percentage points.1 Unfortunately, this narrowing is not due to an increase in cell phone response rates (which consistently have remained in the neighborhood of 10 percent to 20 percent), but rather, the change has come from a further decline in the RDD landline rates. Given that the emphasis in this report is on cell phones, no discussion is presented about why the trend in landline rates continues to decline, although it is hoped that the discussion on cell phone survey nonresponse will, nevertheless, shed some light on declining RDD landline rates.

Two distinct dimensions of nonresponse in cell phone surveys are addressed in this section:

  • What the sources of nonresponse in RDD cell phone surveys are; and
  • How the AAPOR telephone response rate formulae can be modified to account for the unique features of cell phone interviewing. 

Sources of Nonresponse

The reasons for low response rates in RDD cell phone surveys involve the same essential components that account for nonresponse in RDD landline surveys  – noncontact, refusals, other noninterviews and undetermined eligibility. However, these components play somewhat different roles and have somewhat different impacts on overall nonresponse in an RDD cell phone survey compared to an RDD landline survey (cf. Steeh and Piekarski, 2008).

Noncontacts. When considering only noncontacted numbers confirmed as working (i.e., ones that ring but never have been answered by an actual person), the empirical evidence to date suggests that they make up approximately the same proportion of final dispositions in both RDD landline and RDD cell phone surveys provided the number of call attempts is sufficiently large (more than 5) and varied in terms of time of day and day of week. 

The tendency for cell phone owners to constantly carry their cell phones with them and keep them turned on all the time means that they are potentially accessible to interviewers in a much wider variety of settings and for a greater part of their waking hours than in a landline survey. In the coming years, as more people think of the cell phone as their primary phone, the noncontact component of nonresponse in RDD cell phone surveying is expected to decrease. For the same reasons, however, the noncontact component may well increase in RDD landline surveys.

Refusals. Refusals are a main source of nonresponse in RDD cell phone surveys as they are in RDD landline surveys, especially when those surveys carry out many callbacks. However, in several comparison studies of both types of RDD surveys, the refusal rate in the cell phone survey exceeded the rate in the comparable landline survey by five to 20 percentage points. 

Given the structure of the telephone system in the U.S., it is easy to understand why refusals are more numerous when the mode of contact is a cell phone. Reasons for this include:

1.   The called party often is charged for a cell phone call. Even though U.S. service providers now offer a number of different calling plans that provide varying levels of “free” minutes, many potential cell phone respondents will incur costs and so will be likely to refuse immediately when a stranger (the interviewer) contacts them. This problem is exacerbated in U.S. RDD cell phone surveying because potential respondents cannot be “warmed up” with advance mailings containing some form of remuneration as can be done with RDD landline numbers that are matched to household addresses.

2.   The variety of settings in which a cell phone owner might receive a call generates refusals. It has been consistently reported that approximately one third of cell phone survey interviews are completed with someone who is away from her/his home (e.g., Lavrakas, Tompson and Benford, 2010). However, no one has reported the proportion of cell phone respondents reached away from home and thus no one has reported the response rate among those sampled via their cell phone that are reached away from home versus those reached on their cell phone while at home.

However, logic suggests that if a potential respondent is in a restaurant, driving a car, or in a place that is very noisy or crowded, the request for an interview might be met with a hasty “No” and an immediate disconnect before the interviewer even has a chance to mention the possibility of rescheduling the call. 

3.  Some people treat their cell phones as a private device while others use them only occasionally. In the past, many owners appeared to regard the cell phone as a private and personal form of communication shared only with family and close friends, if at all, and a proportion used their cell phones only for outgoing calls and essentially never received or answered incoming calls. There is empirical evidence (as well as myriad anecdotal experience) that supports this observation. In response to an open-ended question about whether they would mind being called on their cell phones by a research organization, respondents in a 2003 survey frequently mentioned invasion of privacy as a primary reason for being opposed (Steeh, 2003). As one respondent replied, “my [cell] phone is for ‘personal use,’ not for annoying people to call me on.”  Thus an interviewer’s cold call may produce an immediate, flat-out refusal. 

This attitude, however, may not be as common in 2010 now that cell phones have become widespread within the adult population. Nevertheless, U.S. cellular numbers still are not listed in any public directory, attesting to the continued pervasiveness of the thinking that the cell phone provides “private” communication.

4.   It is more difficult to convert refusals when the mode of administration is a cell phone. This is because interviewers trying a second call to a previously refusing cell phone number are likely to reach the same person who initially refused, rather than someone else within the household who may be more willing to cooperate (as often happens when conducting refusal conversions in an RDD landline survey).

Taking all of these factors into account, one is led to the conclusion that high refusal rates – and thus low response rates – will plague RDD surveys of persons in the U.S. contacted via their cell phone for the foreseeable future. However, in time the adverse reaction to an interviewer’s call is anticipated to become less intense. Although U.S. service providers have begun to reduce the costs associated with receiving calls on a cell phone, movement to date in that direction has been slow. Were this rate of decreasing cost for receiving incoming calls to accelerate, refusals are likely to decline. In addition, eventually survey methodologists should be able to devise more effective refusal avoidance strategies that are better targeted at sampled cell phones owners, thereby better preventing refusals in the first moments of contact.2  

Nevertheless, although some of the conditions that foster high refusal rates in cell phone surveys may ameliorate over time, refusals are expected to remain substantial because RDD cell phone surveys are exposed not only to unique circumstances, but also to the same negative influences that have been driving down response rates in RDD landline surveys during the past two decades.  

Other Noninterviews. This component of nonresponse consists of two types of failure to achieve cooperation:

1.  The intended respondent cannot physically or mentally participate in an interview, speaks a different language from the interviewer, and/or will not be available throughout the survey field period. As noted previously, empirical evidence shows that higher proportions of non-English respondents are reached in a U.S. RDD cell phone sample than in a U.S. RDD landline sample.

2.  The intended respondent has other reasons for not being able to participate at the specific time(s) when an interviewer calls.

With the exception of language barriers, nonresponse due to these factors has remained low in RDD cell phone surveys and is basically equivalent to that observed in RDD landline surveys provided the level of effort includes multiple callbacks.   In the future, the “other noninterview” component is expected to remain relatively constant and continue to be the smallest and least problematic component of nonresponse in U.S. RDD cell phone surveying.

Undetermined Eligibility. In both cell phone and landline RDD telephone surveys, a proportion of the selected sample ends in an ambiguous region between definitely working and definitely not working, but even when numbers can be classified as working, uncertainty often remains as to whether they are residential numbers. At the end of the field period these numbers are given a final status of “unknown” or “undetermined” eligibility. 

The size of the unknown eligibility component of nonresponse is likely to be much larger in an RDD cell phone survey than in an RDD landline survey because cell phone numbers are not as easily identified as being definitely ineligible for the following reasons:

1.   The plethora of operator messages in the U.S., which differ by provider, often are unclear and confusing to interpret accurately. As a result, it often is very difficult for an interviewer to accurately determine whether, in fact, the number is truly ineligible. Table 2 presents some examples of common operator messages that are highly ambiguous as to whether a cell phone number is working or not working. 



Table 2. Examples of Ambiguous Cell Phone Operator Messages


This phone’s voice mail has not been set up yet.


The number or code you dialed is incorrect. Please check the number or code and try again.


The cellular phone you have called is turned off or out of the service area; please try your call again.


This number is not accepting calls at this time.


Press 0 to speak with an operator.


Please enter the extension of the party you are trying to reach.


Future developments may help to reduce the unknown eligibility problem in U.S. RDD cell phone surveys. The industry is consolidating, and with fewer U.S. companies the jumble of operator messages eventually should result in clearer and more standardized wording. Furthermore, the sporadic use of cell phones should continue to decline as more individuals rely on the technology. Additional research will clarify the direction of the effects of telephone technologies on nonresponse making it easier for survey methodologists to devise effective solutions. For these reasons, the component of nonresponse due to an inability to determine a number’s eligibility is expected to have less and less impact on an RDD cell phone survey’s overall response rate.


Ineligibility in Cell Phone Surveys

When accurately identified and coded, ineligible cell phone numbers do not constitute a form of nonresponse and thus do not lower response rates. However, when they are inaccurately identified and/or coded, they can contribute to various types of error, including making response rates inaccurate.

The following are thought to be the most prevalent types of ineligibility encountered in cell phone surveying of the U.S. general population. Thus, it is important that researchers devise effective means of identifying these ineligible numbers and train interviewers accordingly.

Out of Geographic Area. Researchers need to consider the geographic implications of reaching a cell phone user in light of the target population that the survey is meant to represent. Geographic screening of those reached on their cell phone is necessary in all cell phone surveys that are not national in scope and do not have subnational geographic stratification in their design. 

Unfortunately, geographic screening via telephone often is not easy to carry out accurately. If the wording used for the geographic screening is not well crafted by researchers and well implemented by the interviewers there will be a nonnegligible number of Errors of Omission – false negatives in which someone is incorrectly screened out when they are in fact geographically eligible – and Errors of Commission – false positives in which someone is incorrectly screened in when in fact they are geographically ineligible. Furthermore, screening may add to the number of refusals that occur during the survey introduction when such screening is likely to be carried out, especially if it is not devised to work as parsimoniously as possible.

Age Ineligible Minors. Due to the large number of persons using cell phones who are under the age of majority in the U.S. (i.e., 18 years old in most cases, although some states set the age at 19 or 21), researchers need to establish minimum age requirements for a general population survey calling cell phone numbers, including who can serve as a household informant during the survey introduction. Scripts and disposition codes must be devised for interviewers to use whenever a person under the age of majority is reached. Furthermore, there is anecdotal information that some minors purposely will answer age screeners inaccurately in order to participate in a survey, especially if the survey topic interests them. Researchers should anticipate this possibility in light of the frequency of reaching minors in any U.S. cell phone sample and decide what steps, if any, should be taken to try to minimize this from happening.

Purely Business Cell Phones. Clear and consistent rules for interviewers to determine when a number should be assigned an ineligible disposition of “business phone” should be established. Many persons with a company-provided cell phone use the phone to take both business-related and personal telephone calls (many, but not all, U.S. employers allow this), but they typically do not volunteer that the phone is used for business purposes unless they are asked directly. An a priori decision about whether such numbers should be considered as being eligible in a survey needs to be made by the researchers at the time of planning the survey, and rules should be established for interviewers to accurately determine whether these numbers are eligible or ineligible. For example, an answer of “Anna’s Cleaners” clearly identifies a cell phone used for business purposes; but whether it is used for business only and thus clearly ineligible for a general population survey remains to be determined by the interviewer.

Group Quarters. Traditionally in landline surveys, persons living in group quarters were not considered to be eligible. As cell phones have become commonplace, the affordability of cell phone service has enabled many in group quarters to have their own personal telephone which would not have been possible if it were a personal landline. Thus, survey designers need to consider whether a person reached on a cell phone in group quarters or other persons living in group quarters (e.g., dormitories, military barracks, etc.) along with a sampled cell phone owner should be made eligible sample members. If others in the group quarters are not eligible, then interviewers need to understand that it only is the person initially reached via the cell phone who is the respondent. If others in the group quarters who may also use the cell phone are eligible, then some form of systematic within-unit selection must be chosen by the researcher and deployed by the interviewer. The latter would cause the cell phone to be “handed off” to someone else in some portion of these cases, and this known to lower response rates compared to interviewing only the person who initially answers the cell phone. 

The traditional rule for making anyone reached in group quarters ineligible may hold depending on the researcher’s definition of an eligible housing unit. These decisions defining ineligibility should take into account the fact that it appears that a larger percentage of owners share their phones than has been originally assumed by many researchers. This is especially true when an owner has more than one cell phone.


Nonresponse Bias

Although response rates for RDD cell phone samples tend to be low (and comparable or slightly lower than for parallel landline samples), this does not imply that their estimates necessarily suffer from nonresponse bias. As with landline surveys the level of bias resulting from nonresponse is different for each survey estimate. The bias will be negligible, unless there is a relationship between the likelihood of participating and the outcome of interest – a bias persists after weighting.

Few studies to date have addressed nonresponse bias in cell phone surveys. As a consequence, an understanding of when such relationships are likely to exist is woefully incomplete. However, one can use findings from research studies that have investigated correlates of the propensity to cooperate in a cell phone survey and then make reasonable projections to circumstances under which nonignorable nonresponse bias may well exist in U.S. cell phone surveys.

Research has shown that people interested in technology are more likely to participate in cell phone surveys than those who are not (Brick et al., 2006). Similarly, people who use their cell phones frequently are more likely to participate than infrequent users (Brick et al., 2006). In 2009, the proportion of completed interviews with cell phone only respondents in U.S. national cell phone surveys tended to be twice (or more) as high as the estimated percentage of cell phone only adults within the general adult population. This suggests that (1) contact rates are substantially higher for cell phone only respondents compared to cell phone respondents with landlines, perhaps because they are more likely to answer unidentified calls and/or (2) refusal rates are substantially lower for cell phone only respondents than for cell phone respondents with landlines, perhaps because they are more likely to view their cell phone as a broad access point rather than a private line. This has important implications for the efficiency of sampling cell phone only adults in cell phone samples as well as for the representativeness of unweighted samples from cell phone surveys. For example, studies relying on cell phone samples to estimate levels of interest in technology (or related constructs) or cell phone usage would be expected to overestimate the parameters of interest.    

Additional research is needed to identify other variable domains that are at serious risk of nonresponse bias in cell phone surveys. In particular, it would be useful to know if common correlates of response propensity in landline surveys (e.g., education, age, minority status, civic engagement) are also correlated with response propensity in cell phone surveys. This research could be conducted under different weighting protocols to determine if nonignorable bias remains after such adjustments. 


Outcome Codes and Response Rates in U.S. Cell Phone Surveys

Disposition Codes Used in Cell Phone Surveys. In many instances, the disposition codes and the formulae for calculating response rates published in the AAPOR Standard Definitions for RDD landline surveys can be reasonably well adapted to be applied to RDD cell phone surveys. Some differences lie in the interim (temporary) codes assigned prior to the final disposition and are due to the nature of the call to a cell phone. Other differences are due to new outcome codes that are only possible with cell phone surveys, to current AAPOR landline codes that change in meaning or prevalence when processing RDD cell phone samples, and to codes that apply only to landline surveying and can be eliminated altogether for cell phone surveys (cf. Callegaro, Steeh, Buskirk, Vehovar, Kuusela and Piekarski, 2007; Barron, Khare and Zhen, 2008). 

      Other new disposition codes are required for the situations that do not arise in landline surveys. For example, unlike a landline, a cell phone can be in a geographic area (i.e., depending on the comprehensiveness of the service provider’s geographic coverage) or other location without service coverage (e.g., inside a tunnel; in the basement of a parking garage; etc.). Cell phones also are switched off more often than landline ringers are turned off within a household. Usually in these cell phone instances an operator message or a voice mail message will allow interviewers to classify the outcome into new disposition codes that account for these circumstances (e.g., “not in service at this time”). Although this cell phone survey outcome may appear to be the close equivalent to a “disconnected” or “not working” outcome in a landline survey, its implication is quite different in the cell phone survey. In the landline survey this outcome likely will be a final disposition indicating a number that is ineligible for a completion. In contrast, in a cell phone survey, the disposition is an interim one and the number should be redialed as often as the calling rules for the sample allow in an attempt to reach someone when the phone is turned back on or comes back into a geographic area or location in which its service is working.

      Another situation that does not arise in landline surveys stems from the fact that people may be doing almost anything when they answer their cell phone — walking in a crowded place, driving a car, visiting a restroom, flying in a helicopter, attending a basketball game, eating in a noisy restaurant, etc. (cf. Richtel, 2010). The use of new interim disposition codes that identify these outcomes would help researchers determine how large an effect this “temporary unavailability” has on the processing of cell phone samples.

2.   Codes with different meanings in cell phone surveys compared to their meaning in landline surveys. When calling cell phones, some landline disposition codes may have different or expanded meanings. The “breakoff” code in landline surveys is a case in point. It indicates that the landline respondent herself/himself actively has terminated the interview prematurely. In a cell phone survey, on the other hand, a “breakoff” may also occur as the result of a dropped call or other technical problems and may have nothing to do with the respondent actively deciding to break off from the interview. These kinds of new meanings should be recognized as new interim disposition codes in cell phone surveys – i.e., cases that need different handling than traditional breakoff refusals in landline surveying.

3.   Codes with different prevalence in cell phone vs. landline surveys. The Household Level Refusal code (i.e., a refusal that occurs before a designated respondent has been selected) is an example of a code that has a markedly different prevalence in cell phone surveys. Due to the personal nature of most cell phones in the U.S., a refusal by somebody other than the designated respondent is much less likely to occur in cell phone surveys than in landline surveys. Other examples of codes that arise less often in cell phone surveys include fax machine and busy signals.

4.   Landline codes not applicable in cell phone surveys. The codes in landline surveys that are often not applicable in a cell phone survey include other household level codes, such as group quarters and household level language problems provided the designated respondent is the person who answered the phone.

5.   Ineligible codes and cell phone survey response rates. The AAPOR Standard Definitions guidelines make note of a number of differences in calculating response rates within a cell frame compared to a landline frame (AAPOR, 2009):

(a)  Although in most general population landline surveys researchers can reasonably assume that almost all sample records that are deemed to be a household are eligible to be interviewed, researchers cannot as readily assume eligibility status in cell phone samples because of:
(i)   the geographic area of a specific target population;

(ii)  required screening of specific cell phone users such as those that are cell-only and cell-mostly; and/or

(iii) the likelihood that the cell phone is owned/used by someone under the age of 18 (in studies that must select an adult member of a household ages 18 and older).

In all three of these cases, researchers should only include persons screened as eligible into Category 2 of AAPOR response rate calculations, per the AAPOR Standard Definitions.

(b) In studies that only sample cell phones, it is proper to treat numbers that have been ported to a landline as ineligible for the survey (code 4.46, landline).

(c) Cell phones are more likely to reach owners who are ineligible for many general population surveys, e.g., college students living in dorms, foreign visitors staying in hotels, or day laborers living outside the United States.

6.   Special considerations in calculating response rates for cell phone sample dual frame surveys. In calculating a single response rate for a dual frame survey, it is necessary first to calculate separate response rates for the individual frames. In doing this, there are two issues that should be taken into account in calculating the cell phone rate that do not apply to the landline rate.

First, operator messages tend to be less standardized for cell phones than for landline telephones and vary by company. Some of them are ambiguous or unclear, and thus open to differing interpretations (as shown in Table 2). If this problem is not resolved, the number of unknown eligible cases will remain higher than in a comparable landline survey.

Second, at present it is unclear how the Unknown Eligibility category of nonresponse should be adjusted in Formulae 3 and 4 of the AAPOR Standard Definitions when the mode of administration is a cell phone. How the component, “e” (the estimated proportion of cases of unknown eligibility that are treated as eligible), is defined will have important effects upon response rates. Thus researchers are reminded that if they used these formulae (i.e., 3 and 4), it is incumbent upon them to explicitly disclose the basis on which they formed their estimates of e.

Because of these and other differences between outcome dispositions in landline and cell phone surveying, considerable caution should be exercised whenever rates for single frame cell phone surveys are compared to response rates for landline surveys.

Examples of Calculating Response Rates for Dual Frame Surveys. In dual frame designs, the rates for the units that are sampled from each frame should be combined using weights that are proportional to each segment of the population sampled from the respective frame. 

To illustrate this, two examples of dual frame telephone surveys are considered below.



Previous: Coverage and Sampling
for RDD Cell Phone Surveys

Table of Contents

Next: Measurement in
RDD Cell Phone Surveys

  1. Many business cellular numbers also are used for personal communication and this dual use makes them eligible sample units. Since a telephone number in a general population survey must be used only for business or commercial purposes to be considered ineligible, few cell numbers will meet this standard. Distinguishing between residential and commercial numbers becomes more problematic and less reliable in RDD cell phone surveys than in RDD landline surveys. Without asking, there is no easy method for determining whether a given cell number has reached an eligible respondent on a ”personal” or a “business” cell phone.
  2. Some owners do not turn their cell phones on for long periods of time. They may turn them on only in emergencies or only when they want to make an outbound call. However, and as suggested earlier, this practice, which leaves the working status (and thus the eligibility) of a great many RDD cell phone numbers in doubt at the end of the field period, is definitely on the decline. Current data from a U.S. national study estimate that 87 percent of cell phone only and 81 percent of dual users keep their cell phones on all the time (Carley-Baxter et al., 2010).
  3. Technologies such as answering machines, voice mail, and Caller ID operate differently in cell phone and landline surveys creating uncertainty about eligibility. In telephone surveys of the general population, household answering machines, voice mail and Caller ID are used to identify eligible landline and cell phone numbers. However, their impact on nonresponse may differ for the following reasons: 
    1. Not all landlines are attached to answering machines or have voice mail. Nearly all cell phones have voice mail. Although this may seem to give cell phones an advantage since having the interviewer leave a message when no one answers appears to increase participation (cf. Benford, Lavrakas, Tompson and Fleury, 2010), there are other factors that must be taken into account.
    2. It is unclear whether messages left by the telephones’ owners are qualitatively different on landlines versus cell phones. For example, businesses may be more likely to identify themselves clearly as businesses on a landline than on a cell phone. Given the fact that cell phones are more often used for both personal and business purposes, as noted previously, lends support to this possibility. Furthermore, on a landline there are presumably more instances for leaving a “household” message (e.g., you have reached “the Smiths”) given its inherent nature as a household device rather than a personal device. Thus the content of answering machine and voice mail messages seem likely to be more effective at identifying eligible telephone numbers in landline as opposed to cell phone surveys.
    3. Caller ID has two possible effects:
      1. First, it can provide information about the eligibility of the number. Only about three out of five landlines have Caller ID. Landline Caller ID will display both a number and textual information. On cell phones, however, all non-blocked incoming calls show a number, but give no textual information, unless the incoming call is listed in the recipient’s address book. This means that more information about the eligibility of the number can be gathered from Caller ID as it operates on landlines than compared to how it operates on cell phones. 
      2. Second, Caller ID provides potential respondents with advance information that may influence whether or not they answer the call. Are people more or less likely to pick up an incoming call when they see a strange number on a cell phone versus on a landline?  Although the added text on the Caller ID of a landline has been found to help in raising RDD landline survey response rates, it does not appear to raise the likelihood that a contact will be made (Trussell and Lavrakas, 2005). Whatever the result, it is more likely to impact cell phone rather than landline surveys because every cell phone has Caller ID but every landline does not.
    1. New interim codes. In calls to cell phones, possible new interim codes include “respondent not reachable at this moment” or when a “network busy” message is encountered. When the field period for a cell phone sample is closed, codes such as these that are still in their interim status must be classified into a final disposition status.
    • Example 1: Cell Phone Only Sample and Landline Sample. Suppose a sample of landline telephone numbers is selected from the landline frame and a sample of cell phone numbers is selected from the cell phone frame and screened so that interviewing is done only for the cell-only households. The rates for each of these samples should be computed as noted above for the appropriate single frame design. Special care should be taken to examine the eligibility rate of cell-only households from that frame. Suppose the landline population accounts for 70 percent of the total population and the cell-only population accounts for the remaining 30 percent. In this case, the weighted overall response rate is the sum of 0.7 times the landline response rate and 0.3 times the cell-only response rate. Notice that in this case the weighting factors 0.7 and 0.3 are based on the proportions of the population that are used in the estimates rather than the proportions covered by the frames.
    • Example 2: Cell Phone Sample and Landline Sample with Overlap. Suppose all otherwise eligible cell phone respondents are interviewed (i.e. no screening for cell-only or cell-mostly status). In this case, households with both landlines and cell phones can be reached by either device and thus are called “the overlap.” Also assume it is not possible to identify in advance whether a telephone number from either frame is in the overlap. Assume the landline frame covers 70 percent of the population and the cell frame covers 80 percent, with 30 percent cell only and 50 percent in the overlap. First, compute the proportion of the population sampled from the landline frame as 0.2 (landline only) + 0.5/2.0 (half the overlap) which equals 0.45; for the cell frame it is 0.3 (cell-only) + 0.5/2.0 (half the overlap) which equals 0.55. In this example, the weighted overall response rate is the sum of 0.45 times the landline response rate and 0.55 times the cell frame response rate.

    1 This generalization is based on the response rates for 13 dual frame surveys conducted by the Pew Research Center over the period from 2008 - 2010. All response rates were calculated using the AAPOR Response Rate 3. 

    2 For example, methodological changes, such as leaving a voice message on the first call if no one answers (Benford, Lavrakas, Tompson and Fleury, 2010), may help reduce refusals.  See also further discussion of refusal conversion efforts in the Operational Issues section of this report.   

    3 Fuchs and Busse (2010) have found this to be true in Germany when both “active” and “passive” sharing is measured. Their research shows that asking about whether a cell phone is shared with someone else is a more complex construct to measure than many had assumed.