This study sought to replicate the finding that watching a one-time advertisement for a company or product increases the likelihood of recommending that product or company to a friend. The analysis constructed a latent variable from three net promotor items and tested whether watching a relevant or irrelevant ad embedded in a TV show would increase the latent likelihood. All datasets were opened at 1:15pm on April 19, 2018. A number of demographics were taken as well, including ethnicity, education, income, region, sex, age, and whether the participants were Hispanic. Dummy variables were created for ethnicity (White, Black, Other), education (less than high school, high school grad, some college, college grad), income (less than 30k, 30,000-49,999 , 50,000-75,000, 75,000-99,999, or more than 100,000), region (South, Northeast, Midwest, West), and age (24-34, 34-44, 44-54, 54-64). For all analyses white, collegegrad, less30k, south, and whether they were over 64 years old were the reference categories. Participants were randomly assigned to be invited to the study as part of the first batch of 750 or the second batch of 750. We analyzed the 2nd 750 first, followed by the 1st 750, then combined the two into an overall analysis. 2nd 750: Data collection started at 2:20pm on April 19, 2018 Replication model The replication model represents the same model as was run by the originating lab. Before analysis, the following variables had to be dropped because they had no variance (femalemiss, agemiss, hispanicmiss, racemiss, edumiss). This was done because there was no missing data so these variables were all 0. We created a latent variable representing the netpromotor score for the three variables (How likely are you to recommend McDonald's to a friend or colleague? [metric variable], How likely are you to recommend McDonald's french fries to a friend or a colleague?, and How likely are you to recommend McDonald's food to a friend or a colleague?) and tested whether watching a McDonald’s ad (versus an unrelated ad) caused an increase in latent net promotor scores. This was done while also conditioning on the following variables: sex, whether they were Hispanic, race (White, Black, Other), Education (less than high school, high school grad, some college, college grad), income (less than 30k, 30,000-49,999 , 50,000-75,000, 75,000-99,999, or more than 100,000), region (South, Northeast, Midwest, West), and age (24-34, 34-44, 44-54, 54-64), with missingness indicators for income and region. We did not find any statistically significant differences between those who watched an Ad for McDonald’s versus if they had watched an ad for an unrelated product (b = .02, p > .42). This result was obtained in the full replication model with the necessary changes noted in the 2nd 750 participants who were collected. To calculate a Cohen’s d effect size, we used the estimated mean difference between the groups as the unstandardized b coefficient and divided by the square root of the estimated variance from the diagonal cell of the latent variable of the covariance matrix (outputted using TECH4 in MPlus). This led to an effect size of d = .058, with 95%CI = -.084 to .196. The only demographic variables that significantly correlated with net promotor score were the following: females were less likely to promote McDonalds (b = -.05, p < .049, 95%CI = -.1 to -.001); younger participants in the following age brackets were more likely to recommend McDonalds: 18-24yo (b = .123, p < 029, 95%CI = .013 to .233), 25-34yo (b = .137, p < 001, 95%CI = .058 to .216), 35-44yo (b = .146, p < 001, 95%CI = .064 to .229); Hispanic and Black participants were more likely to recommend McDonalds than White participants (bHispanic = .143, p < 003, 95%CI = .05 to .236), (bBlack = .182, p < 001, 95%CI = .096 to .268); and High School grads were more likely to recommend McDonalds than college graduates (b = .084, p < 011, 95%CI = .019 to .149). Thus, in the second 750, we failed to replicate the finding that watching an ad about McDonalds causes an increase in net promotor score. 1st 750 Data analysis began at 3:16pm on April 19, 2018. Replication model The replication model represents the same model as was run by the originating lab. Before analysis, the following variables had to be dropped because they had no variance (femalemiss, agemiss, hispanicmiss, racemiss, edumiss, incomemiss). This was done because there was no missing data so these variables were all 0. The analysis proceeded as in the 2nd 750 with the minor changes noted above. In the first 750 participants, watching an Ad for McDonalds caused an increase in the latent net promotor score than if they had watched an ad for an unrelated product (b = .103, p < .001, 95%CI = .053 to .152; d = .294, 95%CI = .151 to .433). This effect was significantly larger in the 1st 750 than in the 2nd 750 (bMcD*1st750 = .088, p < .012, 95%CI = .019 to .157). The only demographic variables that significantly correlated with net promotor score were the following: younger participants in the following age brackets were more likely to recommend McDonalds: 25-34yo (b = .113, p < 005, 95%CI = .035 to .192), 35-44yo (b = .185, p < 001, 95%CI = .098 to .272), 45-54yo (b = .137, p < 001, 95%CI = .061 to .213); non-Black non-Hispanic participants were more likely to recommend McDonalds than White participants (bOther = .107, p < 035, 95%CI = .007 to .206); and those who hadn’t graduated from High School (b = .212, p < 018, 95%CI = .036 to .387) and High School grads (b = .098, p < 005, 95%CI = .029 to .166) were more likely to recommend McDonalds than college graduates. Thus, contrary to the 2nd 750 participants, we were able to replicate the effect of watching an Ad increasing the latent net promotor score for that company or product in the 1st 750 participants. Full 1500 Data analysis began at 3:33pm on April 19, 2018 Replication model The replication model represents the same model as was run by the originating lab. Before analysis, the following variables had to be dropped because they had no variance (femalemiss, agemiss, hispanicmiss, racemiss, edumiss). This was done because there was no missing data so these variables were all 0. Otherwise, the full replication model was run. In the full 1500 participants, watching an Ad for McDonalds caused an increase in net promotor score than if they had watched an ad for an unrelated product (b = .061, p < .001, 95%CI = .026 to .095; d = .175, 95%CI = .075 to .273). The only demographic variables that significantly correlated with net promotor score were the following: females were less likely to promote McDonalds (b = -.041, p < .026, 95%CI = -.005 to -.076); younger participants in the following age brackets were more likely to recommend McDonalds: 18-24yo (b = .117, p < 014, 95%CI = .024 to .209), 25-34yo (b = .131, p < 002, 95%CI = .075 to .187), 35-44yo (b = .167, p < 001, 95%CI = .107 to .226), 45-54yo = (b = .102, p < 001, 95%CI = .045 to .159); White participants were less likelty to recommend McDonalds than Hispanic participants (bHispanic = .112, p < 002, 95%CI = .042 to .181), Black participants (bBlack = .114, p < 001, 95%CI = .053 to .175), and all other ethnicities (bOther = .087, p < 012, 95%CI = .019 to .154); and college graduates were less likely to recommend McDonalds than all other levels of education (blessHS = .185, p < 005, 95%CI = .057 to .313; bHSgrad = .098, p < 001, 95%CI = .051 to .145; (bSomeCollege = .056, p < 019, 95%CI = .009 to .095). Thus, using the replication model, we were able to replicate the effect that watching an Ad for a product increases the net promotor score for that company and its products. Follow-up analyses: Confirming Factor Structure We confirmed that the three net promotor items all load onto a single factor that exhibits excellent fit (CFI = 1, RMSEA = 0, SRMR = 0). Thus, we continued to use the latent variable approach in the rest of our additional investigations. ATE model We next calculated a simple average treatment effect (ATE) model with the latent variable of net promotor score and only the treatment indicator of whether participants saw the McDonalds ad or the unrelated Prudential ad (no covariates). This model showed nearly identical effects as the replication model, with participants who saw the McDonalds ad showing a greater net promotor score than if they had seen the unrelated ad (bATE = .058, p < .001, 95%CI = .022 to .094; d = .167, 95%CI = .063 to .27). Robust Treatment and covariate model A number of statisticians have warned about the dangers of mixing randomization with non-randomized covariates in the same analysis (e.g. Freedman, 2008; Lin, 2009; Wager et al., 2016). One proposed solution is to include robust standard errors as well as all treatment by covariate interactions (Lin, 2009). Thus, as the inclusion of a number of demographic covariates was part of the replication model, we fit a model with the treatment effect, the covariates, and all treatment by covariate interactions using robust maximum likelihood estimation. This analysis showed no treatment by covariate interactions (all ps > .17). The treatment effect, however, was no longer significant when including the treatment by covariate interactions (bTxCV = .017, p > .79). This combination of loss of treatment effect from ATE without statistically significant moderation may suggest that there may be effect heterogeneity, but we cannot necessarily detect whether or know it's there (or, at the very least, we cannot detect all of it; see Leeb & Potscher, 2005). Overall results The results from the different analyses led to the same basic conclusion: watching an advertisement about a company or product leads to an increase in the overall likelihood of recommending that product or company to your friends. This was most confidently seen in the ATE analysis that did not involve any additional parameters and provides an unbiased average treatment effect. That the treatment results changed between the covariate included and the robust covariate by treatment interaction analyses suggests that there is heterogeneity in treatment effects but it likely does not vary across the demographics we investigated here.