Willingness of the Public to Share Geolocation Information in a U.S. Census Bureau Survey

Abstract

In 2016, the U.S. Census Bureau conducted a split-panel experiment to explore the public’s willingness to share geolocation information within a survey. A sample of participants from a nonprobability panel were invited to take part in an online survey using their mobile device. Within the survey, one question asked for their address and then the survey requested permission to access their geolocation information. Depending on the study condition, the survey varied how the geolocation request was made and where in the survey the address and geolocation requests appeared. Results showed that the treatment that explicitly asked for permission in addition to the device’s default permission request increased female respondents’ sharing of that data but not male respondents’ sharing. Results also showed that placing the address and geolocation request toward the end of the survey significantly increased the willingness of all respondents to share their location information. Results indicated that respondents with more education and nonminority respondents were more willing to share their location data, but willingness to share location data did not depend on age of the respondent. Assuming that the respondents reported truthfully that they were at home while taking the survey and entered their home address, we found the geolocation data to be accurate to the correct block a little more than 50% of the time.

Keywords

geolocation survey methodology online surveys U.S. Census Bureau

There are many online applications that request a mobile-device users’ geographic location, most notably the mapping and traffic apps using global positioning system (GPS). This geographic location information, henceforth referred to as geolocation, include latitude and longitude coordinates that identify where the device is located at that moment. The World Wide Web Consortium (W3C) is an international community developing web standards. A subset of the group is the W3C geolocation application program interface (API) that is attempting to standardize geographical location information. According to the W3C’s (Popescu, 2016) geolocation API specifications, the application must ask for a user’s permission to collect geolocation coordinates. Consequently, when a website is requesting that information from a device, typically a pop-up message will appear on the screen asking whether the user wants to share the location with the website. When the message is asked, how it is asked, where the individual is at the time it is asked, who is asking, along with the users’ personal privacy concerns and beliefs may play into whether the user will allow the website to know their geolocation (Crawford, 2017; Crawford, McClain, Nelson, & Young, 2014; John, Acquisti, & Loewenstein, 2011; Sadeh et al., 2009; Toch et al., 2010).

In the context of an online census or other data collection that needs to assign people to a particular location, geolocation information might prove very useful as it would, in theory, pinpoint where the person answered the request if the person was using a device with geolocation services on. Alternatively, it could hinder a data collection operation if respondents preferred not to reveal their geolocation or worse yet, perceived the request as invasive and did not participate in the data collection at all. Investigating how to optimize the geolocation request and studying the impact of that request on respondent behavior and survey answers was the rationale for this research.

The U.S. Census and Location Information

The U.S. Census is an address-based enumeration. That is, dwelling units are identified and individuals living or staying within those dwelling units are enumerated (counted). The U.S. Census Bureau maintains a Master Address File (MAF). The MAF aims to include every dwelling unit in the United States. The Census Bureau continually updates and adds to the MAF throughout the decade leading to the decennial census. In 2020, the Census Bureau will rely on the U.S. Post Office to deliver census mail to each housing unit on the MAF. Those census letters and postcards request the recipient to participate in the census. Households will be able to report online, by paper, or on the telephone.

In order to associate an individual with an address, the current design of the online census form asks the respondent to enter an identification number (user ID) located in the mail delivered to their address. Each user ID is associated with an address. Once the user ID is entered into the online form, the Census Bureau knows which address has responded to the census and knows where to count the people enumerated within that online form. If a respondent does not have their user ID, they can still complete the census but the respondent must first provide the address where they were living or staying as of April 1, 2020. The online form then performs an internal check of that address against an extract of the MAF. The matching of the typed addresses to already existing address records is conducted so that the census form collects as accurate address information as possible; however, it is not foolproof. This internal check might not catch all errors. It falls short in two specific situations.

First, an address could be entered incorrectly by the respondent, yet be an actual address on the address reference file. This might happen if a respondent transposes or miskeys their house number but the house number entered, in fact, exists on the MAF. In this case, the internal check would not catch the error. Miskeying on smartphones is actually quite likely. In a recent usability study, 17 of the 29 total participants made errors (e.g., touching the wrong letter or number) as they filled out their address on a mobile phone. In the lab setting where the test was taking place, all but 2 of the 17 participants went back to make corrections—but it is possible that outside of a lab setting, more respondents would continue without making correction(s).

Second, the MAF could be missing a U.S. address. This might happen if the address the respondent entered either was a recently constructed residence or was recently converted from a business into residential housing. In this situation, the internal check would not recognize the address that the respondent typed into the form.

While there are no current plans for geolocation checks in the 2020 Census, these checks could provide additional information that would better facilitate the confirmation of the address’s existence and location. If a respondent were to answer the address question at their home, on a device with geolocation services on, and the home address they typed was confirmed by the geolocation—then, in theory, the address was correct. If there was a keying error, as described above in the first situation, the coordinates could be used to perform a double check because the geolocation would not match the address entered. For the second situation where the address file was not up to date, the coordinates could confirm a new address and possibly be enough information to update the address reference file. However, these possible outcomes assume that the geolocation coordinates themselves are accurate and this is not always the case. Crawford and colleagues (2014) found geolocation to be more accurate and reliable on mobile devices than on PCs. Other researchers have found that determining the users’ geolocation is not as precise when only using the computers’ Internet Protocol (IP) address (Liu et al., 2014).

Current Research Study

Assuming the accuracy of the geolocation information is confirmed, it appears that geolocation may be one way to address the varied challenges of collecting self-reported address information. To this end, we were interested in learning whether respondents would allow the Census Bureau to collect the geolocation coordinates from their smartphone (or any device with WiFi or GPS). While there has been some work on how to ask for geolocation information (Crawford, 2017), we took this opportunity to further study the topic of how best to ask respondents for their geolocation information in an empirical research study. As well, because our study collected geolocation coordinates, self-reported addresses, and the device used to answer the survey, we were able to study the accuracy of the coordinates by device.

We planned and implemented a 2 × 2 split-panel survey experiment with two experimental factors: (1) how the geolocation request is asked and (2) location in the survey where the geolocation and address questions are made. Within each factor, there were two conditions: (1) with and without an overt question in the survey in addition to the default device permission question and (2) placing the address and geolocation request at the beginning versus at the end of the survey. The design was fully crossed. The rest of this article highlights prior research on this topic, our research questions, the methodology, and findings from this experiment.

Prior Research

Prior research suggests the public may have privacy concerns related to geographic tracking (Junglas & Spitzmuller, 2005; Kaasinen, 2003; Minch, 2004). Individuals were more comfortable with geographic tracking when given the ability to explicitly choose which applications were allowed to learn their location (see Gruteser & Liu, 2004; Huang, Matsuura, Yamane, & Sezaki, 2005). In addition, privacy concerns over the sharing of geolocation information seem to diminish somewhat in a student population when overt questions, asking respondents whether they were willing to share their geolocation information, were included. These overt questions were in addition to the phone default “pop-up” where the web application asks to know the current location offering a “Don’t Allow” or “OK” as the response choices (Crawford, 2017).

While there has been some work on whether individuals would be willing to provide their geolocation to federal agencies for emergency situations (Aloudat, Michael, Chen, & Al-Debei, 2014), relatively little research has investigated a respondents’ willingness to provide geolocation information to the Census Bureau. One such experiment investigated the trade-offs associated with geolocation requests versus respondents’ willingness to provide personal information within the survey (Brandimarte & Acquisti, 2014). The researchers found individuals were less likely to share sensitive information about themselves when they knew geolocation information was being collected. This finding suggests individuals could consider location and address information itself sensitive or a barrier to sharing other sensitive behaviors. The Brandimarte and Acquisti (2014) research included sensitive-behavior survey questions where they asked about such things as lying, stealing, and sexual misconduct. The decennial census includes basic demographic questions, such as name, sex, date of birth, race, and ethnicity. In the current research, we were interested in seeing how respondents reacted to geolocation requests with questions similar to what is asked in the decennial census.

Like name and date of birth, the address question is very specific and could potentially be perceived as “sensitive” or “intrusive” due to the fact that after providing this information, the respondent no longer has anonymity (see, e.g., the work by Richman, Kiesler, Weisband, & Drasgow, 1999). Tourangeau, Rips, and Rasinski (2000) have grouped sensitive questions into three broader groups: questions with a “social desirability” element (e.g., when the respondent answers in a way that may be viewed favorably by others), questions with a perceived “threat of disclosure” (e.g., a teen answering about a topic that parents would disprove of), or a question simply being too intrusive. Literature has shown that the context of when the sensitive questions are asked impacts whether the respondent will continue to answer the survey or will opt to break off (i.e., exit the survey without finishing), or refuse to answer (see, for instance, the extensive literature on social desirability bias; DeMaio, 1984; Holbrook & Krosnick, 2010; Kreuter, Presser, & Tourangeau, 2008; Krumpal, 2013; Tourangeau & Yan, 2007).

Oftentimes, respondents may prefer not to answer or change their response to the answer based on social desirability impulses. Strategies that survey designers employ to reduce such a bias include allowing a respondent to answer the sensitive question anonymously and confidentially such as by using a more private mode of self-administered data collection. Strategies also include being aware of question order when designing the survey by paying attention to the questions that are asked prior to the sensitive question and reordering if necessary (Acquisti, John & Loewenstein, 2012; Tourangeau & Yan, 2007). Other strategies include allowing active nondisclosure options (such as allowing an answer option of: “I prefer not to answer”; Joinson, Woodley, & Reips, 2007). But, what about in the case of a question that is sensitive because the respondent perceives it as too intrusive? Perhaps, the solution is contextual as well—by putting questions that are both anonymous and innocuous toward the beginning of the survey and the more intrusive question(s) toward the end. We thought it possible that the placement of when the address question and the geolocation request occurred in the survey may impact respondent answers.

Research Questions

Our research was focused on investigating the following questions:

Were respondents more or less likely to provide their geolocation when an overt question asking for permission was used (in addition to the default question the phone uses) compared with only the default phone question? The overt question also contained a statement “this will help us make sure you are counted in the right place” on the same screen asking for permission to access the location coordinates. Crawford (2017) found that an overt question asking for geolocation coordinates increased a participant’s willingness to allow the coordinates to be shared. To confirm this finding using a different population, we explored this with our Census Bureau nonprobability opt-in research panel.

Did placing the address question (a potentially intrusive question) either toward the beginning or the end of the survey impact participant behavior: Whether they broke off or whether they allowed their geolocation to be shared? In addition to understanding the most effective way to collect geolocation information, answering this question would help us better understand the impact of asking for address data and geolocation information on survey responses. This would further the work of Brandimarte and Acquisti (2014), with respect to the decennial census specific survey questions.

Does the willingness to share geolocation coordinates differ by the demographic characteristics of respondents? This would further the work by Crawford et al. (2014), by examining criteria that might contribute to allowing the geolocation coordinates to be shared.

What is the accuracy of the location information provided? Does it match the address provided? Does the accuracy depend on the type of device provided? (as suggested by Crawford et al., 2014).

Research Methodology

For the sample, we selected 2,000 e-mails from the Census Bureau’s nonprobability affinity panel. The panel consists of people who have opted-in to participate in research studies with the Census Bureau. Of the 2,000 e-mails selected, 1,285 of them had never been selected for a Census Bureau research study, and 715 had responded to an earlier Census Bureau research study and indicated that they owned a smartphone. All the questions in the survey were pretested with six participants in a cognitive test that focused on the question wording.

The study took place over a period of two weeks, May 16–28, 2016. The study provided no monetary incentive to participate. The respondents whose e-mail addresses had been selected in the sample were notified of the study with up to three e-mail invitations requesting their participation in the study. Everyone in the sample received the initial e-mail. The follow-up e-mails were sent only to those who had not yet clicked on the survey link. In the e-mails, we encouraged respondents to answer the survey at home and on their smartphone/GPS-enabled device because we anticipated that those devices should have the geolocation technology built into them.

The message in the e-mails asked recipients to click on the survey link in the e-mail. This led them to a secure, mobile-friendly survey. The username needed to enter the survey was the e-mail address where the invitation e-mail was sent. That username directed the sampled respondent into the assigned survey panel. The survey itself took about 5 min to complete. It asked a series of opinion and demographic questions about the respondent, including their residential address and the assigned geolocation request sequence. The order of the questions and the number of geolocation permission questions varied by the study design, as described below.

The survey panels differed by the factors that might affect a respondent’s willingness to share geolocation information in a Census Bureau online survey. The study had two factors each with two conditions and was fully crossed so that there were 500 e-mails selected for each cell. The first factor dealt with the amount of communication about the geolocation information request:

The control condition for the first factor included three set questions: the residential address (Figures 1 and 2), if they were home (Figure 3), and if they were home, the phone’s default geolocation request that popped up on their screen (see Figure 4).

The experimental condition included those same three set questions and an additional overt geolocation request question. The order of questions for the experimental condition included the residential address (Figures 1 and 2), if they were home (Figure 3), and if they were home, then an overt question within the online survey asking whether the Census Bureau could access their geolocation coordinates (Figure 5). This was followed by the default geolocation request asking for permission (Figure 6). The default geolocation request only appeared if the respondent had answered “yes” to the explicit question. If the respondent answered the explicit question with a “no” then the default phone request never came up. The explicit survey question also gave additional rationale for why geolocation was needed (e.g., “This will help us make sure you are counted in the right place”; see Figures 5 and 6).

Figure 1.

Address question.

Figure 2.

Address question scrolled down.

Figure 3.

“Are you answering from home?”—Question asked of everyone.

Figure 4.

Example of phone’s default geolocation request as it appeared on an Android device (control condition).

Figure 5.

The explicit question asking permission to collect the device’s location (experimental condition).

Figure 6.

Phone’s default geolocation request pops up. On left, as it appears on an Android device, on right with iPhone (experimental condition).

In the second factor, we manipulated the place in the survey that the address and geolocation request occured:

In the control condition, the address question, followed by the geolocation request, was asked at the beginning of the survey, prior to any demographic or opinion questions.

In the experimental condition, the demographic questions came first along with the opinion questions. Then, at the end of the survey, we placed the address question followed by the geolocation request. (See Figures 7 –13 for the demographic and opinion questions).

Figure 7.

Demographic questions.

Figure 8.

Demographic questions scrolled down.

Figure 9.

Demographic questions scrolled down.

Figure 10.

Demographic questions scrolled down.

Figure 11.

Demographic questions scrolled down.

Figure 12.

Opinion questions.

Figure 13.

Opinion questions scrolled down.

Because address data were critical to the evaluation, the survey ended prematurely if the respondent was unwilling to answer the residential address question (Figures 1 and 2). Likewise, if the respondents reported that they were not at home (in Figure 3), the geolocation question(s) were not triggered because we would not be able to study the accuracy of the geolocation coordinates. However, the survey did not end prematurely if they were not home. The respondent simply proceeded to the next question in the survey without seeing the geolocation request(s).

In theory, if the respondent gave his or her approval to share the geolocation information, then the online survey passively captured the latitude and longitude coordinates. However, the survey had no control over how the respondent had previously set up their location data sharing on their phone. Location sharing can be turned off in the “settings” section of a phone. If location sharing is turned off, the phone owner may never receive the pop-up window with the default question about whether they would like to share their location data. Therefore, a situation could occur in this experiment where the respondent indicates they are at home, and reports “yes” to the overt sharing question, but never receives the default question (because they have their phone location turned off), and therefore never share their location data in the survey.

In addition to the address and geolocation information, the survey collected self-reported demographic characteristics of the respondent. It also passively collected the type of device the respondent used to answer the survey.

Analysis and Results

Responses Used

Of the 2,000 e-mails selected, 327 logged into the survey. Of those, 301 submitted responses to the survey; resulting in a response rate for the study of 15%, removing a handful of undeliverable e-mails from the denominator. Of the 301 submitted surveys, only 157 (or 56%) of respondents reported being at home, but there was no difference in the number of responses by survey treatment (χ² = 0.94, p = .33). Only 66% of these respondents followed the instructions to use a mobile device. However, we found that Windows operating systems provided geolocation coordinates at the same rate as mobile operating systems (χ² = 0.31, p = .58); thus, we included in the analysis the 157 respondents who indicated they were answering from home, regardless of the device they used. The demographics of these 157 respondents are found in Table 1.

Table 1.

Sample Characteristics of 157 Respondents Who Reported Being at Home When Taking the Survey.

Device
Mobile	104
PC	53
Sex
Male	54
Female	100
Missing	3
Race
White non-Hispanic	123
Other	34
Education
Less than a college degree	26
College degree or more	129
Missing	2
Age
Less than 34 years old	18
35–44	28
45–54	31
55–64	32
65+	46
Missing	2
Answered a Census Bureau research study in the past…
New sample (never contacted for a Census Bureau research study before)….	120
	37

Source: May 2016 Small-Scale Nonprobability Internet Geolocation Test (n = 157).

General Findings

To answer the first three research questions, we ran a logistic regression model predicting whether location coordinates were collected (1 = coordinates were collected; 0 = coordinates were not collected). In the model, we controlled for the type of device (PC or mobile); the typical characteristics of the respondent including race and ethnicity, age, education, and sex; whether the respondent had answered a previous research survey; and the two experimental factors: (1) amount of communication about the geolocation information request and (2) placement within the survey of the address question/geolocation request. We ran several models with interaction terms. We found no significant interaction between the two experimental factors. There was, however, a significant interaction between the sex of the respondent and the amount of communication about the geolocation information request. Table 2 contains the estimates, standard errors, and the odds ratios of this model. Overall, the experiment showed that we collected coordinates from 52 (or 33%) of the 157 people. Results are discussed using α = .05 as significant and anything between .05 and .10 as marginally significant.

Table 2.

Logistic Regression Model Predicting the Likelihood of Collecting Geolocation Coordinates.

Independent Variable	Estimate	Standard Error	Odds Ratio	p Value
Communication factor: asking an overt question and the default phone question to collect coordinates versus not asking an overt question but only the default phone question	.22	.20	Not provided	.27
Place in the survey factor: placing the address and geolocation question(s) at the end of the survey versus early in the survey	.52	.21	2.83	.01
Mobile device versus PC	.23	.21	1.58	.28
Sex: Male versus Female	.14	.21	Not provided	.49
Race: White non-Hispanic versus all other	.46	.26	2.50	.08
Education: Some college or less versus college degree or more	−.63	.31	0.28	.04
Age
<34 years old versus 65+	−.61	.50	0.53	.22
35–44 versus 65+	−.51	.45	0.59	.26
45–54 versus 65+	.64	.40	1.85	.11
55–64 versus 65+	.45	.38	1.5	.23
Was never selected for a previous study versus answered a previous study	−.16	.26	0.73	.54
Communication factor* sex	−.67	.21	Not provided	.002

Source: May 2016 Small-Scale Nonprobability Internet Geolocation Test (n = 154)*.

*Three observations were dropped from the model because they had missing demographic characteristics.

To answer the fourth research question, determining the accuracy of the geolocation coordinates, we used data from the 52 respondents who allowed their coordinates to be collected. Census Bureau geography staff matched the latitude and longitude coordinates obtained to the geocode coordinates based on the addresses provided by the respondent. Five codes were assigned based on the analysis:

Code 1: the latitude and longitude coordinates were an “exact match”

Code 2: not an “exact match,” but in the same block

Code 3: in an adjacent block

Code 4: close by (adjacent to the adjacent block)

Code 5: not close by (which is anything farther than the fourth code)

Because it is extremely unlikely to get the exact coordinates, here “exact match” means the location of a point in which we have compared the respondent provided point to a geocoded point (based on the given address) and that falls within a 2,600 square foot proximity. The 2,600 square foot proximity is based on the average size of a home. Table 3 contains the results of the coding of the 52 home addresses with geolocation coordinates. We used χ² statistics for analysis.

Table 3.

Accuracy of the Geolocation Information Collected Compared to the Geocode of the Address Collected by Device Type.

Match Code	Smartphone	PC	Total Across Device	Percentage
Exact match (within 2,600 sq. ft.)	5	4	12	23.1
Same block	12	4	16	30.8
Adjacent block	13	3	16	30.8
Close by (adjacent to the adjacent block)	1	0	1	1.9
Not close by (anything else)	2	5	7	13.5
Total	36	16	52	100

Source: May 2016 Small-Scale Nonprobability Internet Geolocation Test (n = 52).

Research Question 1

Were respondents more or less likely to provide their geolocation when an overt question asking for permission (in addition to the default phone question) was used compared with only the default phone question. Model results shown in Table 2 indicate that there was a significant interaction between the overt question’s use and the collection of geolocation coordinates by sex. When the overt question and the default phone questions were used, there was no difference in the collection of geolocation coordinates by sex (χ² = 0.41, p = .52): In this condition, coordinates were collected from 33% of males and 40% of female respondents. However, when the overt question was not used and only the phone’s default question was used, coordinates were collected from only 19% of female respondents compared to 46% of male respondents which was significantly different (χ² = 5.6, p = .002).

Research Question 2

Did placing the address question either toward the beginning or the end of the survey impact participant behavior: Whether they broke off or whether they allowed their geolocation to be shared? Because 327 people logged into the survey, but only 301 submitted data, we know that 26 broke-off within the survey. We conducted a χ² test of independence and found that break-offs were more likely to occur in the condition where the address and geolocation request occurred toward the beginning of the survey (χ² = 5.2, p = .02), but breakoffs were not impacted by whether there was an overt question (χ² = 0.18, p = .67).

Not only did break-offs decrease if the address and geolocation questions were at the end of the survey, but the model results in Table 2 show that placing the address and geolocation request at the end of the survey increased the collection of location coordinates compared to when the address and geolocation request were at the beginning of the survey (p = .01). The odds of sharing location information were 2 times greater when these data were asked at the end of the survey. For this study, only 26% of respondents were willing to share their geolocation information when the request was made early in the survey compared with 41% who were willing to share it when the request was made at the end of the survey.

Research Question 3

Do the demographic characteristics of respondents help predict their willingness to share location data? Based on the model results in Table 2, we found:

Respondents without a college degree were less willing to share their location data than respondents with a college degree (p = .04).

White non-Hispanic respondents were marginally more willing to share their location data than non-White respondents (p = .08), and

There was no detectable difference at the 95% confidence level in the willingness to share location data by age, type of device used, or by whether they had previously participated in a Census Bureau research study.

Research Question 4

What is the accuracy of the location information provided? Does it match the address provided? Does the accuracy depend on the type of device? Based on the results in Table 3, we found:

A little over 53% of the geolocation coordinates would code the response to the correct block.

Only 13.5% of the geolocation coordinates were far from the correct block while the remaining 32.7% were either in an adjacent block or close by.

There was no detectable difference at the 95% confidence level in the accuracy of the geolocation to the correct block by device (χ² = 0.14, p = .71); however, PCs were more likely to be “not close by” than smartphones (χ² = 6.3, p = .01).

Discussion

We collected geolocation coordinates from 33% of our respondents, indicating that at this time, one third of the public may be willing to divulge their location coordinates with the Census Bureau—and thus allowing an easy way for the Census Bureau to validate those addresses. We did not find any indication that asking an additional overt question about obtaining location coordinates in addition to the default phone question increases break-offs. Obtaining geolocation data from respondents that are willing to share it could contribute to better data collection operations overall. While it is true that the other 67% may consider the geolocation request invasive or private information that they do not feel comfortable sharing, there are still the traditional strategies that the Census can use to validate their addresses. And, for those who shared their location, a little more than half of the coordinates provided would map the survey response to the correct block, which would be necessary if we were to use these data to assign survey responses to a geographic location.

We found female respondents less willing to share their location data with only the phone’s default permission question compared to men. But, by adding the overt request for the participants’ geolocation along with a rationale for why Census needed that information, we were able to eliminate the difference in sharing between men and women. Crawford et al. (2014) found an effect with such a question with a student population’s willingness to share location information but does not find any difference between men and women. The two populations are different demographically (e.g., Crawford et al. (2014) used university students; our study used a nonprobability panel of participants who had indicated some level of interest in the Census Bureau). Still, for both populations, it could be that asking for permission (and providing a rationale) prior to having the default phone request makes any participant who agrees feel like they are giving a commitment to proceed. In psychological testing, those that make a commitment to do something are more likely to carry through and do it (see, for instance, the work by Greenwald, Carnot, Beach, & Young, 1987, on voting). Research also shows that making a commitment to provide accurate information yielded more accurate information (e.g., Cannell, Oksenberg, & Converse, 1977; Conrad, 2011). It is possible that the online geolocation request is tapping into what has been uncovered in prior commitment studies.

We found that when the address question came at the beginning of the survey, participants were less likely to allow sharing of their geolocation coordinates than when the address question and the geolocation request came at the end of the survey. We also found more break-offs to the survey when the address and geolocation requests were at the beginning of the survey. Both of these findings suggest that address and geolocation information is sensitive and correct placement of sensitive questions is critical, as other researchers have demonstrated (Acquisti et al., 2012; Holbrook & Krosnick, 2010; Tourangeau & Yan, 2007).

In our study, we also found that sharing geolocation information depended upon the characteristics of the respondent. Respondents without a college degree were less willing to share their location data than respondents with a college degree and White non-Hispanic respondents were marginally more willing to share their location data than non-White respondents. There was no detectable difference at the 95% confidence level in the willingness to share location data by age of the respondent or the type of device he or she used. The fact that some of the respondents had reported to an earlier Census Bureau survey did not affect whether or not they allowed their location data to be shared. It is possible that the demographic differences with respect to sharing geolocation can be attributed, in some extent, to privacy concerns (see, e.g., O’Neil, 2001; Singer, Mathiowetz, & Couper, 1993), although more research on how requests for geolocation impact survey responses by demographics is needed.

These findings contribute to the literature and confirm some of the previous work. Implementing and using geolocation coordinates to associate individuals or households with a specific location still needs additional research. We learned that sharing the coordinates is not universally accepted, and there are currently technical glitches that make sharing difficult, even when the respondent is willing (i.e., having the location services off). As the technology evolves, the approach may become more precise and should be considered a possibility for survey administrators.

Limitations and Future Work

The population for this experimental study is a sample of people who have opted-in to participate in Census Bureau research studies. These individuals could be more willing to provide their geolocation information than the general public because they trusted the Census Bureau enough to provide their e-mail when they initially signed up for the opt-in panel. Future research should investigate a different set of participants—specifically to include those that are not connected to the Census Bureau in any way. Additionally, such research will add to the body of evidence on how asking for geolocation information in a survey impacts response rates and the respondents’ willingness to share geolocation information. Additional experiments could look into response rates when collecting address information at the beginning of the survey and seeing the impact of waiting until the end of the survey to collect the GPS coordinates, or vice versa.

Finally, we close with some thoughts on the situation where participants who intend to provide their geolocation information but are unable to do so because they happen to have their location services turned off. We assume the percentage of respondents who have the location settings turned off is the same in both panels due to the random assignment of conditions, but because we could not measure whether our assumption is true, it becomes a confounding factor. It is also possible to imagine that some respondents really intended to give permission but didn’t realize that having location services turned off made that impossible. Future research might suggest a work around to such a situation, such as in identifying whether the phone is preset to deny all location requests and, if so, prompting the respondent with information on how to change this in real time.

Footnotes

Authors’ Note

The Census data used in our research are confidential and protected by Title 13 and Title 26 of the U.S. Code. The U.S. Census Bureau supports external researchers’ use of these data for approved projects with valid statistical purposes through the Federal Statistical Research Data Centers administered by Center for Economic Studies (see http://www.census.gov/fsrdc). There are more than a dozen Federal Statistical Research Data Centers (FSRDCs) located at universities and research institutions across the country. For information on the locations, please see . Access for purposes of replicating a scientific study is a valid statistical purpose. Please contact Erica L. Olmsted-Hawala at Erica.l.olmsted.hawala@census.gov for guidance on getting started with accessing this data. We will archive our data and make it available for use for approved projects at the FSRDCs. We will also provide the programs that generated our statistical results. This report is released to inform interested parties of research and to encourage discussion. The views expressed are those of the authors and not necessarily those of the U.S. Census Bureau. The authors would like to thank Sarah Dietz and Glenn Eanes who created and managed the survey; Christopher Jackman, Nicholas Doner, and Edwin Brightmon for their geocoding; Frank McPhillips for sponsoring this work; Sabin Lakhe for his help with parsing the paradata; and Joanne Pascale, Lin Wang, Benjamin Reist, Frank McPhillips, Dan Donello, Paul Beatty and our reviewers for their reviews of this article.

Declaration of Conflicting Interests

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: The authors worked for the Census Bureau while they worked on this project.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

References

Acquisti

John

Loewenstein

(2012). The impact of relative standards on the propensity to disclose. Journal of Marketing Research, 49, 160–174.

Aloudat

Michael

Chen

Al-Debi

(2014). Social acceptance of location–based mobile government services for emergency management. Telematics and Informatics, 31, 153–171.

Brandimarte

Acquisti

(2014). Aiming at a more cost-efficiency census via online data collection: Privacy trade-offs of geo-location. Presented at the annual NCRN meetings at the Census Bureau, Suitland, MD, May 22–23, 2014.

Cannell

Oksenberg

Converse

(1977). Experiments in interviewing techniques: Field experiments in health reporting, 1971–1977. Hyattsville, MD: National Center for Health Services Research.

Conrad

. (2011, 5). Interactive interventions in web surveys can increase response accuracy. In Proceedings of the American Association for Public Opinion Research (AAPOR) 66th Annual Conference. Phoenix, AZ.

Crawford

(2017, 4). Consent form design for geolocation capture in mobile web surveys. In Proceedings of Federal Computer Assisted Survey Information Collection Conference (FEDCasic). Suitland, MD.

Crawford

McClain

Nelson

Young

III . (2014, 5). Understanding mobility: Consent and capture of geolocation data in web surveys. In Proceedings of the American Association for Public Opinion Research (AAPOR) 69th Annual Conference. Anaheim, CA.

DeMaio

Theresa J.

(1984). Social desirability and survey measurement: A Review. In Turner

C. F.

Martin

(Eds.), Surveying subjective phenomena (Vol. 2, pp. 257–82) New York, NY: Russell Sage Foundation.

Greenwald

A. G.

Carnot

C. G.

Beach

Young

(1987). Increasing voting behavior by asking people if they expect to vote. Journal of Applied Psychology, 72, 315–318.

10.

Gruteser

Liu

(2004). Protecting privacy in continuous location-tracking applications. Security & Privacy, IEEE, 2, 28–34.

11.

Holbrook

Krosnick

(2010). Social desirability bias in voter turnout reports: Test using the item count technique. Public Opinion Quarterly, 74, 37–67.

12.

Huang

Matsuura

Yamane

Sezaki

(2005). Enhancing wireless location privacy using silent period. Proceedings of the Wireless Communications and Networking Conference, IEEE, 2, 1187–1192.

13.

John

Acquisti

Loewenstein

(2011). The best of strangers: Context dependent willingness to divulge sensitive information. Journal of Consumer Research, 37, 858–873.

14.

Joinson

A. N.

Woodley

Reips

U. D.

(2007). Personalization, authentication and self-disclosure in self-administered Internet surveys. Computers in Human Behavior, 23, 275–285.

15.

Junglas

I. A.

Spitzmuller

(2005). A research model for studying privacy concerns pertaining to location-based services. In Proceedings of the 38th Annual Hawaii International Conference on System Sciences. Big Island, HI: IEEE.

16.

Kaasinen

(2003). User needs for location-aware mobile services. Personal and Ubiquitous Computing, 7, 70–79.

17.

Kreuter

Presser

Tourangeau

(2008). Social desirability bias in CATI, IVR and web surveys: The effects of mode and question sensitivity. Public Opinion Quarterly, 72, 847–865.

18.

Krumpal

(2013). Determinants of social desirability bias in sensitive surveys: A literature review. Quality & Quantity, 47, 2025–2047.

19.

Liu

Zhang

Zhou

Zhang

Ramakrishnan

K. K.

(2014). Mining checkins from location-sharing services for client-independent IP geolocation. In Conference Proceedings INFOCOM, 2014 Proceedings IEEE. Toronto, Ontario. doi:10.1109/INFOCOM.2014.6847987

20.

Minch

R. P.

(2004). Privacy issues in location-aware mobile devices. In Proceedings of the 37th Annual Hawaii International Conference on System Sciences. Big Island, HI: IEEE.

21.

O’Neil

(2001). Analysis of internet users’ level of online privacy concerns. Social Science Computer Review, 19, 17–31.

22.

Popescu

(Ed.). (2016). Geolocation API Specification 2nd Edition (Technical Report from the W3C). Retrieved from https://www.w3.org/TR/geolocation-API/

23.

Richman

W. L.

Kiesler

Weisband

Drasgow

(1999). A meta-analytic study of social desirability distortion in computer-administered questionnaires, traditional questionnaires, and interview. Journal of Applied Psychology, 84, 754–775.

24.

Sadeh

Hong

Cranor

Fette

Kelley

Prabaker

Rao

(2009). Understanding and capturing people’s privacy policies in a mobile social networking application. Personal and Ubiquitous Computing, 13, 401–412.

25.

Singer

Mathiowetz

Couper

(1993). The Impact of Privacy and confidentiality concerns on survey participation: the case of the 1990 census. Public Opinion Quarterly, 57, 465–482.

26.

Toch

Cranshaw

Drielsma

Tsai

Kelley

Springfield

… Sadeh

. (2010). Empirical models of privacy in location sharing. In Proceedings of the 12th ACSM International Conference on Ubiquitous Computing, ACM (pp. 129–138). Copenhagen, Denmark—September 26–29, 2010.

27.

Tourangeau

Rips

Rasinski

(2000). The psychology of survey response. New York, NY: Cambridge University Press.

28.

Tourangeau

Yan

(2007). Sensitive questions in surveys. Psychological Bulletin, 133, 859–883.