Publication date: 11/30/2021

Author’s note: Product teams at Meta rely on research along with other external factors to design and build products. This article discusses research conducted or referenced by Meta's Research Team to better understand people’s privacy needs.

Abstract

According to Protection Motivation Theory, people are motivated to take protective actions against perceived threats when they believe they can successfully address those threats.

This led us to hypothesize that Messenger users may be most likely to use privacy controls when they have a privacy concern and also have high self-efficacy.

We surveyed Messenger users across eight countries to test this hypothesis, and we found that respondents were most likely to report using Messenger’s end-to-end encryption and app lock controls if they had relevant privacy concerns and high self-efficacy for using Messenger’s privacy controls.

Report

At Messenger, we’ve been developing tools to help people manage their privacy (you can learn about some of those tools here), and we wanted to better understand how we might encourage people to use them.

We began by reviewing academic literature about when and why people engage in behaviors that help them stay safe. According to Protection Motivation Theory (Rogers & Prentice-Dunn, 1997), people are motivated to take protective actions against perceived threats when they believe they can successfully address those threats. There are two important aspects of this theory. The first is the idea of a perceived threat. In the context of online privacy behaviors, people tend to perceive something as being a potential privacy threat if they believe it is likely to occur and also that it would be bad or severe if it did occur (e.g., Milne, Labreque, & Cromer, 2009, Rifon, LaRose, & Lewis, 2008; perceived privacy threats are also sometimes referred to as “privacy concerns”, see Hepler, 2021).

However, according to Protection Motivation Theory, simply perceiving a threat (“being concerned”) isn’t sufficient to cause someone to take protective actions against it. The second important aspect of this theory is whether someone believes they can successfully address the threat that they perceive. One way researchers have thought about this concept is through the lens of self-efficacy, which is a feeling of confidence that you’re capable of doing something successfully (Bandura, 1982). Putting these two aspects of the theory together, academic researchers have found that people are most likely to use privacy controls when two conditions are met: (1) They have a privacy concern and (2) they have high self-efficacy for using an app’s privacy controls (e.g., Milne et al., 2009; Rifon et al., 2008; see a list of related research on privacy self-efficacy here).

Table 1. Theoretical likelihood that someone will use an app’s privacy controls as a function of their privacy concern and self-efficacy.

Table 1

Do self-efficacy and privacy concerns predict use of privacy controls on Messenger?

Based on the academic literature discussed above, we hypothesized that Messenger users would be more likely to use Messenger’s privacy controls if they were concerned about a privacy topic that those controls were meant to address and if they had high self-efficacy for using those controls (see Table 1 above).

We tested this hypothesis by surveying 4,977 Messenger users and measuring (a) privacy concerns related to Messenger, (b) self-efficacy for using Messenger’s privacy controls, and (c) self-reported use of two prominent privacy controls on Messenger: End-to-end encryption and app lock. Here are descriptions of these controls from Messenger’s privacy website.

  • End-to-end encryption: When you want to remain completely private, you can choose to send end-to-end encrypted messages or make end-to-end encrypted calls. That means that messages and calls can only be seen or heard by you and the person you send them to, and no one else - not even us - can listen in.

  • App lock: If you’re looking for more security, opt in to App Lock. Use your device's face or fingerprint ID to unlock Messenger, giving you extra protection for your chats.

For additional methods details including the exact wording of survey questions and additional details about the survey sample, see the Appendix below.

Consistent with our hypothesis and the academic literature, survey respondents who said they had a privacy concern relevant for these controls and who said they had a high level of self-efficacy for using Messenger’s privacy controls were the most likely to self-report that they had used end-to-end encryption and app lock (see Tables 2 and 3 below). In contrast, respondents were less likely to report using these controls if they had (a) a privacy concern but low self-efficacy (b) high self-efficacy but no privacy concern, and (c) no privacy concern and low self-efficacy.

Table 2. The percent of respondents who self-reported using end-to-end-encryption on Messenger, as a function of low and high levels of privacy concern and privacy self-efficacy.

Table 2

Table 3. The percent of respondents who self-reported using app lock on Messenger, as a function of low and high levels of privacy concern and privacy self-efficacy.

Table 3

Therefore, consistent with Protection Motivation Theory (Rogers & Prentice-Dunn, 1997), use of privacy controls on Messenger may depend on users having a combination of high privacy concern and high privacy self-efficacy. In contrast, simply being concerned about privacy might not always be enough to motivate users to take an action like using a privacy control if that concern isn’t paired with enough self-efficacy.

Open questions and opportunities

One important caveat for this research is that the data were correlational. So we can’t determine whether privacy concerns and self-efficacy caused respondents to use end-to-end encryption and app lock, or whether they were correlated for some other reason. Future research could be done to help confidently establish a causal link - for example by increasing users' self-efficacy and determining whether that leads to increased use of privacy controls.

However, if self-efficacy and privacy concerns do cause people to use privacy controls, then there may be opportunities for thinking about how to use these concepts to encourage users to engage with privacy controls on apps like Messenger. 

How might apps help users to have a high level of privacy self-efficacy? This is an open question in need of additional research. However, two promising opportunities for improving self-efficacy are to increase users’ awareness of the privacy controls that are available to them and to make sure those controls are perceived as easy to use. Both of these outcomes might help increase people’s privacy self-efficacy. To that end, Messenger has been exploring ways to increase users’ awareness of its privacy features and to show that they’re easy to use. For example, Messenger launched a privacy website that prominently lists the privacy controls that are available in the app and also provides brief video tutorials showing how to use them.

Can these results help explain the “privacy paradox”? In general, privacy concerns don’t predict how people use products. Across decades of research, this has consistently been found for a wide range of businesses, products, and experiences including social media apps, hardware, and customer loyalty programs (Dienlin & Trepte, 2015; Kokolakis, 2017). The surprising fact that privacy concerns generally don’t predict behavior has been dubbed the “privacy paradox”. However, past research generally hasn’t accounted for users’ privacy self-efficacy. Our results suggest that the privacy paradox might not occur once researchers account for self-efficacy. However, we should note that our research relied on self-reported behaviors (see the Appendix below for details), so it’s important to replicate these results with observed behavior before forming a strong conclusion here.

Summary

In line with Protection Motivation Theory, we found that Messenger users were most likely to report using privacy controls if they were concerned about a privacy topic that those controls were meant to address and if they had high self-efficacy for using those controls. Although these results were correlational, they indicate that a promising path to encourage users to adopt privacy controls is to explore ways to increase users’ self-efficacy related to those controls.

Appendix: Research Methods and Analysis

Study Details

We conducted a survey with Messenger users (N = 4,977, 52% women, average age = 37 years) across 8 countries (United States, Canada, United Kingdom, Germany, France, Poland, India, and Philippines) to understand how the use of privacy controls, privacy concerns, and privacy self-efficacy might be related on Messenger. The survey was translated into the primary languages for all countries in the survey, and respondents were invited to take the survey in the Messenger app via an in-app notification.

We measured privacy concerns using the Privacy Beliefs & Judgments framework (Hepler, 2021). Specifically, we measured concerns about data use and interpersonal privacy (topics that were relevant for the specific privacy controls we explored in the study). Respondents were coded as having a privacy concern if they endorsed both the belief and the judgment items for a given privacy topic (e.g., if they checked the [data use] response options to both questions below then they were coded as being concerned about data use on Messenger).

  • Belief (likelihood): “During the next month, do you believe any of these things could happen to you on Messenger? Select all that you believe could happen to you.”
    • [data use] Messenger will use information from your messages for its business
    • [interpersonal] Someone will see things you've shared on Messenger that you don't want them to see
    • Note: The question included two additional options that were not included in the current report as they did not relate to end-to-end encryption or app lock.
  • Judgment (severity): “If any of these things happened to you on Messenger, would you be concerned? Select all that seems bad to you.”
    • [data use] Messenger used information from your messages for its business
    • [interpersonal] Someone saw things you've shared on Messenger that you didn't want them to see
    • Note: The question included two additional options that were not included in the current report as they did not relate to end-to-end encryption or app lock.

We measured self-reported prior use of end-to-end encryption and app lock using the following question: “Which of these features have you used on Messenger? Select all that you have used before.” The response options included Secret Conversations (encrypted chats), App Lock, and two additional options that didn’t relate to our self-efficacy hypothesis; in the response option list, the name of each feature was accompanied by a brief description of the feature to ensure respondents knew what the feature name referred to. Note that we relied on self-reported use of these features instead of observed use because our hypothesis was that privacy concerns and self-efficacy may predict lifetime use of these features (as opposed to current settings).

We measured self-efficacy using three survey questions. For each question, respondents could answer on a 5-point scale ranging from (1) Not at all confident to (5) Very confident.

  • Global privacy self-efficacy: “How confident are you that you can use Messenger controls to protect your information on the app?”
  • Data use privacy self-efficacy: “How confident are you that you can control how Messenger uses your information?”
  • Interpersonal privacy self-efficacy: “How confident are you that you can control who can see things you have shared on Messenger?”

Results

In the main body of the report, we presented the percent of respondents who self-reported using end-to-end encryption and app lock as a function of privacy concerns and privacy self-efficacy. In addition to simply comparing the means (as we did in Tables 2 and 3 in the report), we also used regression analysis to formally test for a two-way interaction. Specifically, we examined the interactive effect of privacy self-efficacy and privacy concern on self-reported use of privacy controls with the PROCESS approach. Four regression models were run, two predicting use of end-to-end encryption and two predicting use of app lock. In each model, respondent age, gender, and country were included as controls in addition to the relevant privacy concern and the relevant privacy self-efficacy measures. Analyses were rerun replacing the topic-specific privacy self-efficacy measure with the global privacy self-efficacy measure. Three of the four regressions models showed that the interactive effect of privacy concern by privacy self-efficacy was statistically significant (p < .05).

When respondents had low (1 SD below the mean) and medium levels (mean) of privacy self-efficacy, their data use concern was unrelated to self-reported use of end-to-end encryption. But, as respondents’ privacy self-efficacy increased to high levels (1 SD above the mean) the odds that respondents had reported using end-to-end encryption increased significantly. This effect was seen at high levels of both global privacy self-efficacy (B = .59, p = .004) and data use self-efficacy (B = .56, p = .008). The corollary is that privacy self-efficacy only predicted self-reported use of end-to-end encryption when respondents had data use concerns (B = .20, p = .021 for the effect of topic specific self-efficacy and B = .26, p = .004 for the effect of global self-efficacy on end-to-end encryption adoption) but not when they did not have this concern.

When respondents had low privacy self-efficacy (1 SD below the mean), their interpersonal privacy concern was unrelated to self-reported use of app lock. But, as respondents’ privacy self-efficacy increased to medium (mean) and high levels (1 SD above the mean) the odds that respondents reported using app lock also increased significantly (B = .58, p < .001 at medium levels of self-efficacy and B = .91, p < .001 at high levels of self-efficacy). Privacy self-efficacy increased the odds that respondents reported using app lock, but only when they had interpersonal privacy concerns, not if they didn’t. This effect was observed for the global privacy self-efficacy measure (B = .19, p = .047 for the effect of self-efficacy on app lock adoption for respondents who had interpersonal privacy concerns), but not for the interpersonal privacy self-efficacy measure.