Are politically diverse Thanksgiving dinners shorter than politically uniform ones?

Abstract Americans on the political left and right are engaged in a Culture War with one another, one that is often characterized by mutual fear, antipathy, and avoidance. Are there safe havens from the socially straining effects of this Culture War, times and places where Americans of different political stripes gather and put aside their political differences? Previous research (Chen & Rohla, 2018) implied that there might not be insofar as even intimate family gatherings seem to be subject to Culture War tensions. They found that politically diverse Thanksgiving Dinners were 35–70 minutes shorter than politically uniform ones, representing a 14–27% reduction in overall dinner duration. Noting analytical and methodological limitations in the prior analysis, we conducted two pre-registered studies to test whether diverse dinners are shorter than uniform ones and to attempt to conceptually replicate and extend this prior analysis. Individual analyses yielded mixed results, with null models generally supported but effect estimates generally overlapping with those of Chen and Rohla (2018). A mega-analysis found that, when controlling for various covariates, politically diverse dinners were 24 minutes shorter than politically uniform ones, 95% confidence interval = [9, 39], representing a 6% decrease in the total dinner time [2%-10%]. This final result successfully replicates Chen and Rohla (2018) both in terms of effect overlap and direct-and-significance criteria while nonetheless favoring the conclusion that politics is not straining family ties as much as previously thought.

Citation: Frimer JA, Skitka LJ (2020) Are politically diverse Thanksgiving dinners shorter than politically uniform ones? PLoS ONE 15(10): e0239988. https://doi.org/10.1371/journal.pone.0239988 Editor: Valerio Capraro, Middlesex University, UNITED KINGDOM Received: March 12, 2020; Accepted: September 17, 2020; Published: October 27, 2020 Copyright: © 2020 Frimer, Skitka. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: All data are available on the Open Science Framework at https://osf.io/3se4z/. Funding: Social Sciences and Humanities Research Council of Canada (SSHRC) Insight Grant (IG 435-2018-1221) to JAF. https://www.sshrc-crsh.gc.ca/home-accueil-eng.aspx The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist.

Introduction The United States is in the midst of a Culture War [1, 2]. The majority of politically engaged citizens report that they feel afraid of and anger toward members of the other party. This interpartisan animosity appears to be a growing phenomenon; the percentage of partisans who reported “very unfavorable” views of the other side more than doubled in the past two decades, from ~20% in 1994 to ~55% in 2016 [3]. The rift has become so strong that it seems to now be stronger than racial, religious, or ethnic tensions [4]. Both sides are motivated to avoid the other side, even showing a willingness to forgo money to avoid hearing what the other side has to say [5]. These sentiments are strong enough that they can manifest as discrimination at work [6] and in relationships [7]. The result is spatial sorting, with people tuning in to ideologically congenial news outlets, echoing the sentiments of likeminded others in social media bubbles [8, 9], and moving to politically homogenous neighborhoods [10]. The psychological motives that lead to spatial sorting are at least twofold [5]. First, hearing from unlike-minded others means that a person consumes information that conflicts with their existing beliefs. This causes the observer to experience cognitive dissonance, which tends to be accompanied by an aversive feeling [11–13]. Second, agreeable conversations with others about important beliefs and values satisfies a fundamental human need to belong [14] and for a shared sense of reality [15], whereas tense disagreements threaten the same. These theorized and previously empirically supported mechanisms set out the specific conditions that are most likely to induce spatial sorting. Simply being in close proximity to unlike-minded others might not necessarily cause sorting; however, when people talk about ideological and political topics, the social atmosphere should become unpleasant and this sentiment should drive unlike-minded people away from one another. The question that motivates this research concerns whether there are limits to this political rift in the form of times and places where people set aside their political differences and get along for the sake of enjoying an event together. A prime candidate for such a truce might be Thanksgiving dinner, a traditional family gathering in which relatives come together to enjoy a meal and one another’s company, and to feel gratitude for what they have. Norms concerning polite company proscribe conversations about politics (as well as sex and religion), making Thanksgiving a prime candidate for suppressing relationship-straining political conversations in politically diverse company for the sake of getting along. The available evidence, however, suggests that Thanksgiving dinner is not immune to the Culture War between liberals and conservatives. Chen and Rohla [16] analyzed ~25 billion smartphone location data pings of ~10 million Americans in 2015 and 2016 to reach the conclusion that politically diverse Thanksgiving dinners were “30 to 50 minutes shorter” (p. 1020) than politically uniform ones, a substantial 12–19% reduction of the total dinner time (M = 257 minutes). Chen and Rohla’s [16] innovative recruitment of smartphone location data was ground-breaking insofar as it permitted precise measurement of times and locations and generated a massive data set. These features allowed the researchers to detect potentially small but important effects with precision. Without self-reports or behavioral measures attesting to the political leanings of the participants, Chen and Rohla devised a new and innovative approach to infer the political beliefs of dinner participants. This approach drew inferences from the political leaning of the traveler’s home precinct and of the Thanksgiving dinner location. However, we note (and detail below) that the validity of this inference is questionable, leaving open the broader question about whether political diversity really does shorten Thanksgiving dinners and by how much. To understand the questionable features of the political diversity measure requires an understanding of the specific methods that Chen and Rohla used. Political diversity is the phenomenon whereby two or more people hold different, mutually oppositional, political beliefs, and is thus tantamount to a mismatch between the political positions of members of the social gathering. For instance, in a dinner event attended by a Democrat and a Republican, diversity is high whereas a dinner with two Democrats has low diversity. An intuitive approach to operationalizing diversity would involve directly assessing the political attitudes of each person in attendance, by asking them or acquiring their voting record for example, and then computing diversity as the degree of mismatch. Chen and Rohla relied on a different and novel approach to measure political diversity. They inferred the political beliefs of each person from the voting precinct in which their home was located (determined by their location in the middle of the night several days before Thanksgiving) and publicly available information about the general voting tendency in that same precinct in the 2016 U.S. Presidential Election. To understand how this method works and its’ critical limitations that motivate the present work, we present an illustrative, hypothetical example in which a hypothetical person named Joe travels ~25 miles west from his home in downtown Austin, TX (277th precinct) to a Thanksgiving dinner in the western suburb of Austin near Lake Travis (232nd precinct). Knowing nothing more than Joe’s home location and the location of the dinner, the method would infer that the political diversity (probability of a mismatch) of the dinner was .53 on the 0–1 scale. The method begins by inferring that Joe is probably a Democrat by virtue of the fact that his precinct voted in favor of Hillary Clinton over Donald Trump in the 2016 U.S. Presidential election by a margin of 72% to 21% [17]. That is to say that ideological position was operationalized as a probability of being a Democrat (in this case the probability is 77% based on Chen and Rohla’s formula). Meanwhile, the inferred political leaning of a home near Lake Travis in the western suburbs, is computed in the same way and ends up being a 45% probability of being Democratic (The 232nd precinct favored Trump over Clinton by a margin of 52% to 42%; [17]). The mismatch was then computed as a function of these probabilities of 77% and 45% to arrive at a relatively high diversity estimate of .53 (see Chen and Rohla [16], for the formula), which was approximately 1 SD above the M diversity in Chen and Rohla’s sample. In contrast, Joe’s next-door neighbor Alice traveled to a dinner approximately the same distance from downtown but toward the south near Bluff Springs (413rd precinct). Alice’s dinner would have a relatively low diversity score of .33 (~ 1 SD below the M in Chen and Rohla) because like 277th precinct downtown, the 413th delivered 75% of the vote to Clinton and much less (18%) to Trump. We note several limitations with this method. An analytical limitation was that Chen and Rohla [16] extrapolated beyond the populated range of the data to conclude that diversity shortens Thanksgiving dinner by 30–50 minutes, which would mean that the 30–50 minute figure might be an over-estimate. This was accomplished by plotting political diversity (x) against dinner duration (y) and taking the unstandardized slope from a regression analysis as the estimate of the dinner-shortening effect of political diversity. Because the slope reflects the change in dinner duration for every 1-unit change in political diversity, this analytical decision effectively compares the duration of dinners with no diversity (0.00) and dinners with full diversity (1.00). However, very few dinners met these descriptions. The average diversity score across all dinners in Chen and Rohla [16] was .44 with a SD = .10, meaning that ~68% of all dinners had a diversity score between .34 and .54. Chen and Rohla thus drew inferences about the dinner-shortening effect of diversity by effectively comparing dinners that were 4.4 SDs below the M with dinners that were 5.6 SDs above the M, which amounts to an extrapolation beyond the populated range. The 30–50 minute estimate thus described the difference in dinner duration of two hypothetical dinners that virtually never occurred. In personal communication (June 8, 2016), Chen and Rohla suggested that although political diversity was measured using a continuous measure (ranging from 0–1), the event that the measure aimed to model was a dichotomous event: either the two persons at dinner voted for the same presidential candidate in 2016, or they did not. This latent dichotomy would have justified the extrapolation to the theoretical points of interest, namely diversities of 0 and 1. Frimer and Skitka [18] challenged the claim that Thanksgiving dinner diversity is a dichotomous construct, maintaining that political diversity is more accurately thought of along a continuum. The dichotomous claim requires Thanksgiving dinners be a dyadic event, with just two participants, and that both of them are Democrats and/or Republicans. We suggest that neither of the empirical premises are sustainable. The vast majority of Thanksgiving dinners (98% by our estimates) have more than two people present, with an average of 7 people in attendance (see our Studies 1 & 2). Thus, each dinner has some combination of Democrats and Republicans (e.g., 2 Democrats and 5 Republicans), making the amount of political diversity therein fall on a continuum. Moreover, many Americans did not vote for either Trump or Clinton in the 2016 election. Some 58% of the U.S. population and 44% of eligible voters did not vote in 2016 [19], and 6% of voters chose a candidate other than Trump or Clinton. These sizable non- and third-party voter populations represent ambiguous additional categories with respect to political diversity, which further undermine the dichotomous claim that would have justified an extrapolation beyond the populated range of the data. We maintain that combinations of multiple Democrats, Republicans, Independents, and non-voters are more accurately described along a continuum of diversity, rather than in terms of a dichotomy. Accepting that diversity is a continuous construct, and with only Chen and Rohla’s data being available at the time, Frimer and Skitka [18] applied a more conventional procedure for estimating the dinner-shortening effect of political diversity by comparing the duration at M diverity + SD diverity to the duration at M diverity −SD diverity ; doing so produced an estimate of a 4–11 minute reduction from the average dinner time of 257 minutes, a relatively small 2–4% reduction in total dinner duration. Chen and Rohla’s [16] modeling of Thanksgiving dinner as a dichotomous event raises further questions about the validity of the measurement of political diversity. Their data attempted to represent the political leanings of two members of a larger dinner event. For the measure to survive this validity threat, the political diversity observed between the traveler and the dinner homeowner would need to be representative of the diversity of all attendees. This has yet to established. Moreover, measuring the political diversity of a dinner from the home location of the traveler and the homeowner assumes that each traveler’s political leanings are representative of his/her home precinct and that each homeowner’s political leanings are representative of his/hers home precinct as well. This may or may not be true. The additional available information about each person’s Thanksgiving dinner might justify modifications to those assumptions but are not accounted for within the Chen and Rohla formula. The limitations of the existing measure and analyses cast doubt over firm conclusions about whether politically diverse Thanksgiving dinners are shorter than politically uniform ones, and if so by how much.

The present studies Our primary goal was to test whether politically diverse Thanksgiving dinners were shorter than politically uniform ones in the years 2018 and 2019. Our secondary goal was to conceptually replicate Chen and Rohla’s [16] findings, using different and complementary methods. Whereas Chen and Rohla used location data from smart phones to infer the diversity of dinners and their duration, we used a crowdsourcing methodology, asking participants to self-report the time they arrived and departed from Thanksgiving dinner and to tell us about the people in attendance and their political attitudes; we used these data to infer political diversity. The method we employ avoids some of the limitations of Chen and Rohla’s study (detailed previously). Two pre-registered studies, conducted in 2018 and 2019 respectively, measured political diversity and Thanksgiving dinner duration, tested whether the two are associated, and used these data to estimate the dinner-shortening effect associated with political diversity. Do our correlational studies meaningfully replicate Chen and Rohla’s analyses? Chen and Rohla interpreted their fixed-effects regression analyses on their longitudinal, observational data as having established a causal effect of diversity on dinner duration. Our approach was different in that we employed an OLS regression analysis while controlling for various confounds that we measured. Our analyses might not appear to be suitable for testing whether Chen and Rohla’s causal effects replicate. Even noting the methodological differences, we suggest that our correlational observations do offer a meaningful opportunity for a conceptual replication and for two reasons. First, Chen and Rohla’s causal inferences presuppose correlations. Causal effects require that (a) the causal variable is correlated with the effected variable, (b) the causal variable precedes the effected variable, and (c) third variables do not explain the association. If any of the three are unsupported, the causal claim fails [20]. Therefore, our correlational tests have the potential to (dis)confirm the correlational feature of a causal relationship (feature a). Second, we question whether Chen and Rohla’s data justified causal inferences in the first place. Fixed-effects regression analyses allow for causal inference if and only if potential confounds are time invariant between the earlier and later observations (November 2015 to November 2016 in their case). Many important changes took place between Thanksgivings 2015 and 2016, with the most notable being the ascendancy of Donald Trump to the U.S. Presidency. For the time invariance tenet to hold, the psychology of the American electorate would need to have been largely unchanged by the campaign and electoral victory of President Trump. This seems to be a strong and tenuous assumption, raising questions about Chen and Rohla’s causal interpretation. If the time-invariance assumption is relaxed, then the two sets of analyses become conceptually comparable as tests of whether diverse dinners are shorter than uniform ones. Another potential difference between Chen and Rohla’s study and ours might be the time in the election cycle. Chen and Rohla’s main findings were from 2016, a U.S. Presidential election year whereas our tests did not take place on a Presidential election year. However, one of our studies was in 2018, which was a midterm election year. Voter turnout in both 2016 and 2018 was comparable and around 50%-55%, meaning that political engagement was roughly equivalent. Moreover, Chen and Rohla reported that diverse dinners were shorter than uniform ones in both 2015 and 2016 (see their S1 Table in S1 File, models 1 & 2). Thus, we suggest that the years 2018 and 2019 are suitable to testing the effects that Chen and Rohla observed. Our criteria for replication success are twofold. The first is direction and significance: do we observe effects that are in the same direction (and reach significance) as those reported in Chen and Rohla? The second replication criterion is (unstandardized) effect overlap. Chen and Rohla estimated the effect of diversity on dinner duration in one zero-order model and three corrected models (using fixed-effects regression to control for covariates), with the former predicting a 22 minute decrease, 95%CI = [19, 24] and the corrected models predicting 38 [35, 41], 45 [37, 53], and 56 [42,70] minute decreases. We use our zero-order correlations to test whether Chen and Rohla’s zero-order effect replicates and corrected correlations (controlling for covariates) to test whether the latter three replicate. To conservatively test for replication of the corrected estimates, we take the minimum and maximum estimates from the three analyses in aggregate [35, 70] as the target 95% confidence interval. We also test whether our effects replicate Frimer and Skitka’s [18] interpretation of Chen and Rohla’s [16] data, which yield zero-order estimate of a 4–5 minute reduction and a corrected estimate of a 7–14 minute reduction. A secondary goal was to extend Chen and Rohla [16] by proposing and testing a social psychological explanation for the shorter dinners. Chen and Rohla’s [16] explanation for the effect was sociological, implicating the role of political advertisements. Our interpersonal level of analysis led us to theorize that political diversity shortens Thanksgiving dinner because talking about politics induces tension and conflict between un-likeminded people. If so, we should expect that diversity will predict shorter dinners especially when the conversation gravitates to politics (a moderator). The corollary would be that politically diverse gatherings might be able to avoid the socially deleterious effects of diversity by avoiding the topic of politics (as politeness rules suggest). We also examined the theoretically implicated affective mediator, that diversity makes the social atmosphere unpleasant and this negative atmosphere in turn shortened dinners.

Discussion Using a complementary method to that used in previous research, we found little evidence to support the notion that politically diverse dinners were shorter than politically uniform ones, and we found positive support for the conclusion that diverse dinners tend to be of similar duration as politically uniform ones. Bayesian analyses favored the null over alternative hypotheses. These results provisionally support the idea that families are able to set aside their political differences at Thanksgiving dinner, and that political differences are not straining family ties as much as previously thought. The notion that political differences are straining family ties led us to predict that this would be especially the case when the topic of conversation drifted to politics. This only rarely happened. Politics was the least common topic of conversation (sports, work, friends & family, and food were more common topics). But for those dinners that waded into politics, we expected to find political diversity to sharply shorten the dinner. This notion was unsupported, which could mean that talking politics is unpleasant regardless of the attitudes of the conversationalists or that talking politics is not unpleasant enough to shorten a dinner event. Following from the notion that political diversity is straining family ties, we also predicted the diversity would sour the social atmosphere, and this negative social atmosphere would in turn predict shorter dinners. We almost found evidence to support for this idea, with the confidence interval of the indirect effect just barely overlapping with zero. One possible explanation for why social atmosphere (almost) mediated the (null) effect of diversity on dinner length is that some unobserved feature of the social environment was working in opposition to the dinner-shortening effect of a soured social atmosphere. For example, diversity might make the atmosphere less pleasant while also making it more engaging. If people generally like pleasantness and also like engaging atmospheres, it is possible that two competing forces are canceling each other out. In Study 2, we test for this possible suppression effect. Previously reported effects did not replicate using the direction and significance criterion however they did replicate with respect to the effect overlap criterion, a pattern that points toward the possibility that the effect size estimate that we used to conduct our power analysis was too large and that a small effect of diversity on dinner might exist. Although the sample size of Study 1 was theoretically grounded, empirically derived, and pre-registered, it produced somewhat imprecise estimates. Our zero-order analysis found that politically diverse dinners were somewhere between 23 minutes shorter and 30 minutes longer than politically uniform dinners, a result that cannot test Frimer & Skitka’s [18] estimate relative to the null. Ideally, a sample would be large enough for effect size estimates not overlap with the more than one prediction. To achieve this, we would require a 95% confidence interval of 4 minutes. Extrapolating the power curve observed in the present study (see the S1 File), we estimate that a sample of almost half a million participants would be needed, a sample size that is not feasibly achieved via crowdsourcing. The (very) large sample size needed to detect the effect proposed by Chen and Rohla highlights the value of Chen and Rohla’s large data set, which included the smart phone location data of ~10 million Americans. At the same time, it highlights how small in statistical terms the effect they observed was. Our effect size estimate of r ~ .01, 95% confidence interval [-.07, .09] is a small fraction (4%) of the size of the effect that is said to be driving it (selective exposure motivation, r ~ .30). It might well be the case that political diversity shortens Thanksgiving dinners. But, based on the available evidence, it seems unlikely that the effect is an appreciable one. Several methodological limitations of Study 1 are also worth noting. First, when asking about who attended dinner and about their political beliefs, the survey asked about people. Some of the named individuals might have been children without clear (or any) political beliefs. In Study 2, we ask about adults specifically. Second, in Study 1, we measured attitudes only with respect to approval or disapproval of President Trump. Although Trump is a defining figure in contemporary American politics, it would be helpful to include other measures of political attitudes to test whether the effects observed generalize.

Discussion Despite doubling our sample size, Study 2 still failed to produce consistent evidence that political diversity predicted shorter Thanksgiving dinners. Our estimates had very wide confidence interval estimates, which overlapped with both Chen and Rohla’s findings and with Frimer & Skitka’s reinterpretation, meaning that effect overlap replication success was high. However, most Bayesian analyses favored the null over these alternative hypotheses. These results again support the idea that political differences are not straining family ties as much as previously thought. Like in Study 1, we again failed to find evidence of moderation and mediation derived from the notion that political diversity is straining family ties because people dislike hearing from the other side of the Culture War. That said, doubling the sample size increased the replication rate using the direction-and-significance criterion from 0% to 50%, which again points to the possibility that a very small effect still exists. With effect size estimates ranging from r = -.01 to -.09 depending on the operationalization of political diversity, the available evidence suggests that the effect of diversity on dinner duration is unlikely to be an appreciable one. Study 2 allayed several minor methodological concerns raised in Study 1. The inclusion of the degree to which the conversation was engaging mediator, asking about adults rather than about people, and asking about travel time and duration turned out to yield few new insights. However, including three measures of diversity—operationalized with respect to attitudes toward Trump, attitudes toward impeachment, and partisan leaning—did pay dividends. Specifically, when diversity was defined with respect to partisan leaning (but not with respect to President Trump), we found some evidence that diversity predicts a shorter dinner duration when controlling for a raft of contextual factors. It could be that only when diversity is measured with respect to partisan leaning and/or using a larger, more sensitive scale (7-point in this case), and when controlling for contextual factors, does diversity predict dinner duration. Alternatively, this finding could be a one-off (that is, a false positive). Future research should investigate these possibilities. Although not conclusive, the present findings tended to support the null over the claim that diversity predicts Thanksgiving dinner duration.

General discussion Are there safe havens from the socially detrimental effects of the Culture War between liberals and conservatives? Previous research [16] found that even Thanksgiving dinner was not immune to the effects of the Culture War, implying that there may be few safe havens from the conflict. Noting limitations in the previous analysis, we conducted two studies aimed to test whether diverse dinners are shorter than uniform ones and to conceptually replicate and extend the original study. Using a complementary method, as well as pre-registered and well-powered studies, we generally failed to reject the null hypothesis (thus failing to replicate Chen and Rohla by the direct-and-significance criterion), and Bayesian analyses tended to favor the null hypothesis that diversity and Thanksgiving dinner duration are unrelated. However, most effect size estimates overlapped with those in the original, leaving open the possibility that a statistically small effect exists. A mega-analysis of zero-order effects again favored the null. However, controlling for the effect of covariates, a mega-analysis suggested that diverse dinners are 24 minutes shorter than uniform ones, 95%CI = [9, 39], representing a 2%-10% reduction in overall dinner time. This result replicates Chen and Rohla’s [16] conclusions by both criteria. The effect size estimate was r p = -.047, 95%CI = [-.017, .077], which is approximately six times smaller than that of the proposed explanation (selective exposure). A Bayesian analyses weakly supported the alternative hypothesis by 65% to 35% probability. For a replication to be meaningful, theoretically relevant contextual factors need to match those of the original. One possible meaningful difference is change over time. The original study [16] was in 2015 and 2016 whereas the present studies were in 2018 and 2019. If tensions between the two parties had calmed over time, then we might expect political diversity to strain family ties less over time. However, polling data [22] suggest that the rift between the political left and right in the U.S. is not in decline; if anything, it is increasing. Another possibility is that election years bring politics to mind for many Americans, and thus amplifies the family-straining effect of differences of opinion. The original study included a Thanksgiving dinner during a presidential election year (2016) whereas the replication studies did not. However, we note that the effects in the original were also found in 2015, which is at the same time in the electoral cycle as 2019. Moreover, the 2019 study took place in the midst of a highly publicized and politically divisive impeachment of a president, and the 2018 study took place in a midterm election year. These considerations suggest that the context in the replication studies were probably germane to the original claims. A significant difference between the original and the present investigations concern methodology and sample sizes. Whereas the original studies used location data from ~10 million Americans’ smartphones and publicly available election results to measure political diversity and dinner duration, we recruited the self-reports of ~1,500 crowdsourced participants. The sample size differences are substantial and allowed the prior analysis to make more precise estimates and potentially to detect a signal in a (literally and statistically) noisy situation. However, our Bayesian analyses were not equivocal about the null vis-à-vis the alternative. That is, our individual samples were sufficiently large to reach a clear conclusion that politically diverse dinners are no longer or shorter than politically uniform ones, on average. However, our mega-analytic analysis that controlled for covariates favored the opposite conclusion, that diverse dinners are shorter than uniform ones (albeit tentatively so). The methodological differences between the prior and current analyses are substantial, rendering the present effort a conceptual and not a direct replication of the original. Motivating the change in methodology was noted limitations of the original methodology. However, the present methodology was not without its own limitations. For instance, we relied on participants to report the time that they arrived at and departed from the dinner the next day. Before the dinner, we instructed them to record the precise time of arrival and departure. However, only ~60% of participants reported that they were confident about their reported times to within 5-minutes. Including only these participants yielded similar conclusions. However, it remains a possibility that our methodology for measuring the duration of the dinner introduced a sufficient amount of noise to drown out a real signal. Noting the methodological limitations of both the prior and the current approaches to measuring political diversity and Thanksgiving dinner warrants some appreciation of the inherent challenges in studying statistically small effects in private settings. Recognizing these challenges temper our objective to the goal of adding useful information to this important question, rather than to offer the final word. That said, we believe that the bulk of evidence thus far suggests that although people expect conversations with unlike-minded others to be painful [5], they over-estimate the severity of the negative affect of these actual conversations [23]. It appears that the Culture War division that exists is not as strong and toxic as generally thought [24], that many norms surrounding civility and politeness remain intact [25, 26]. With perhaps only a small disruption attributed to politics, Americans appear to be largely successful at putting aside their political differences and enjoying Thanksgiving dinner with relatives and friends with whom their differ.