The Committee for the Protection of Human Subjects at Dartmouth College reviewed the study protocol and granted this study an expedited Institutional Board Review.

To obtain information on doctor jokes posted on Facebook, we used data from the Harris Interactive Research Lifestreaming Panel. Upon agreeing to become a Lifestreaming Panel participant, these individuals gave Harris permission to record their private conversations on social networking sites such as Facebook and Twitter. Although both Facebook and Twitter could potentially be used to study jokes about doctors, we were granted access specifically to the Lifestreaming Panel’s Facebook data. Lifestreaming Panel participants are paid the sum of US $1.00 initially, and they are then eligible to participate in future surveys and activities in which they can receive other payments. Harris collects data from Lifestreaming Panel participants’ social networking sites as well as information on the participants’ characteristics. Harris can search Lifestreaming Panel participants’ Facebook walls using algorithms based on keywords. As of June 2011 (the time of our data collection), there were 33,326 adult Lifestreaming Panel participants.

On Facebook, each user creates a profile and has a personal “wall,” a place for conversing with others. For the purposes of this study, we defined a conversation thread as starting with a root post on a user’s wall by either the user or by the user’s “friend,” and including others’ follow-up comments. A fictional example of a conversation thread in which a doctor joke is mentioned can be found in Figure 1 (note: the authors are used to represent Facebook users to protect the identify of study participants).

Using Harris Interactive Research Lifestreaming Panel Facebook data from December 15, 2010 to June 16, 2011, we identified 30,075 individual posts that included the term “doctor(s)” (Figure 2). Because posts were identified first by the keyword doctor(s) and then attached to full Facebook wall threads, 60 doctor jokes were made by Facebook users not part of the Lifestreaming Panel (these individuals posted on a Lifestreaming Panel participant’s Facebook wall) and were therefore excluded from our study.

To identify jokes, 2 independent coders read each post that included the term doctor(s) in search of jokes. For practical reasons, coders were asked to only identify what they considered to be canned jokes. A typical canned joke contains an introduction followed by a punch line, and can stand alone as a humorous statement, free from context [24]. Coders identified 442 posts that contained potential canned jokes. Differences in coding were reconciled by a third reviewer. Because Cohen’s kappa statistic can underestimate the interreviewer agreement when analyzing rare events, we also calculated the positive and negative agreement between reviewers [25,26]. The kappa statistic, positive, and negative agreement for initial agreement between reviewers was 0.62, 0.65, and 0.99, respectively (Figure 2).

Three study investigators then read all 442 potential jokes and excluded those that were not canned jokes or that contained multiple embedded jokes. Differences were discussed and resolved by consensus. The process resulted in the identification of 321 posted doctor jokes that represented 156 unique canned jokes (some jokes were repeated). Selected examples of the jokes identified can be found in Table 1. We included doctor jokes in our sample whether or not the joke received follow-up posts from the social network.

We used Stata version 13.0 statistical software (StataCorp LP, College Station, TX, USA) for all analyses. The chi-square test for categorical variables and the t test for continuous variables were used to compare sociodemographic characteristics of jokers to nonjokers, except for measures of Facebook activity (ie, social network size and the number of fan and group pages to which a participant belonged), which we compared by using a Mann-Whitney test. Missing values were assigned a category for the comparison of characteristics between jokers and nonjokers for categorical variables.

Because some jokes were repeated in our sample, we used generalized linear mixed-effects models that were fit to allow each joke to have a random intercept. Fixed effects in our models included the characteristics of the joke and the covariates for age, sex, and network size as a categorical variable (ie, small vs medium or large social network). To determine if specific characteristics of jokes predicted the success of a joke, we assumed a binomial distribution for our models in which the dependent variable was whether or not the joke received at least one electronic laugh (yes vs no), and a Poisson distribution for our models in which the total number of likes was the dependent variable.

Among the 33,326 Facebook users in the Lifestreaming Panel, 29.99% (9994/33,326) mentioned the word doctor during the 6-month observation period, but only 263 (0.79%) posted a doctor joke. Jokers varied by US region (P<.001) and were more heavily represented in the Northeast (the region with the highest density of physicians per capita) [27] and the South (Table 2).

Jokers differed from the typical Lifestreaming Panel participant in several ways: they were less educated (eg, 16.0% of jokers reported having graduate training or an advanced degree whereas 20.38% of nonjokers did, P=.02) and they were more likely to be divorced, separated, or widowed (vs married, P<.01). Jokers had larger social networks than nonjokers: jokers had a median of 227 (IQR 138-369) Facebook friends compared to 132 (IQR 56-270) among nonjokers (P<.001).

Of the 156 unique jokes, 112 (71.8%) appeared only once in our data. Of the jokes that were repeated, 2 jokes were particularly popular and were repeated approximately 30 times.

Among the 156 unique doctor jokes, the joke was made at the expense of doctors (and/or the health care system) in approximately half (62/156, 39.7%). In addition, 25.0% (39/156) of jokes relied on dirty humor, 19.9% (31/156) were puns, 14.1% (22/156) pertained to popular culture and media, and 5.8% (9/156) related to current events and politics.

Approximately half of all jokes posted (133/321, 41.4%) received electronic laughter. Jokes made at the expense of doctors were more likely to receive electronic laughter although this did not reach statistical significance (46.5% vs 37.3%, P=.09). The marginal trend (OR 1.46, 95% CI 0.94-2.29) toward a higher rate of electronic laugher in response to jokes at doctors’ expense was only slightly attenuated after adjusting for age, sex, and social network size (Table 3). The likelihood of generating an electronic laugh was not dependent on other joke characteristics in both our univariate and adjusted models.

The median number of Facebook likes for doctor jokes was 2 (IQR 0-19). Ironically, the joke with the greatest number of Facebook likes (49 total likes from a network of 253 friends) was a “doctor, priest, lawyer” joke in which lawyers were the butt of the joke.

We observed similar associations between the characteristics of jokes and receiving Facebook likes as a measure of joke success (Table 3 vs Table 4). Jokes that were at the expense of doctors received nearly 50% more likes (rate ratio [RR] 1.48, 95% CI 0.96-2.27) in our adjusted model (Table 4). However, jokes that used what we deemed dirty humor were actually less likely to receive Facebook likes, whereas jokes that were based on current events/politics received more than double the amount of Facebook likes (RR 2.36, 95% CI 0.97-5.74; P=.06).

To our knowledge, this is the first study to use actual Facebook conversations to examine doctor-related humor. Overall, we found a low prevalence of doctor jokes on Facebook and relatively few Facebook users posting jokes about doctors (and the health care system in general). Interestingly, those who posted a doctor joke were more likely to be divorced, separated, or widowed, and to have larger social networks (ie, more friends on Facebook). Given the previous findings that people who want to improve their social status are more likely to joke, it may be that divorced, separated, or widowed Facebook users tell doctor jokes to appeal to a potential partner but, of course, we cannot distinguish the reason from these data. Although initially it appeared that poking fun at doctors (as compared to doctor jokes that were not made at the expense of physicians) led to more successful jokes (in generating electronic laughs and the total number of Facebook likes), such findings were not statistically significant. In regards to Facebook likes, jokes based on current events and politics appeared to receive greater response from an individual’s social network, whereas dirty humor jokes received fewer likes.

We also observed that although most jokes appeared only once in our data, a few jokes were repeated many times. Based on our qualitative review of these more highly repeated jokes, they differed little from other jokes in our study. Although we are not able to determine the reason, these findings demonstrate how certain ideas can spread rapidly throughout social networks [28].

In recent years, medicine has not been immune to the impact of social networking sites, and there is growing interest in social networking sites among physicians and biomedical researchers. Social networking sites may have important applications for studying social interaction and communication related to health and medicine [29,30]. The medical community has largely focused on discussing the ethics of doctor-patient interaction on social networking sites [31-37] and professionalism of younger practitioners’ exposure via social networking sites [38-44]. Only more recently has interest emerged in using social networking sites to employ health interventions [45,46] and to identify certain health behaviors [47-49]. To date, there have been few empirical studies in the biomedical literature that examined conversations on social networking sites in nonpatient population groups.

Although our study examined doctor jokes posted on Facebook and does not represent a comprehensive analysis of public opinion of the medical profession and health care, our analyses are among the first to examine actual social networking site conversations [50]. Primary analysis of Facebook conversations could provide researchers the ability to examine certain health behaviors and popular opinion pertaining to health and medicine. Furthermore, analysis of social media conversations on a larger scale could have important uses, such as studying US public opinion regarding national health policy, developing new methods for public health surveillance, and for sociological study to understand social support for illness in virtual settings. However, as our study demonstrates, conversations from social media sites contain a mixture of both relevant and (depending on the use) potentially irrelevant material.

Our study has several limitations that must be acknowledged. First, considering that Lifestreaming Panel participants gave permission to have their Facebook data recorded, the potential for selection bias cannot be ruled out. Second, we used the term “doctor(s)” to identify posts pertaining to medical physicians, which may underestimate the total conversations pertaining to physicians, medical practitioners, or the medical profession. Given the casual nature of Facebook, we thought that the term doctor would be used more commonly than a more formal term such as “physician.” Finally, we identified potential jokes for analysis based on our definition of what constituted a joke in Facebook posts; others might have defined jokes differently than we did.

Despite the inherent limitations of our research, this study demonstrates the potential of using social networking sites for research on health and medicine. The adoption of social networking has resulted in growing interest in using outlets such as Facebook and Twitter in creative ways. In this study, we demonstrate how actual data from Facebook conversations can be used to study doctor-related humor. In addition to serving as an example, this study highlights some of the practical considerations regarding the analysis of data from social networking sites.