Norms of assertion in the United States, Germany, and Japan

The recent controversy about misinformation has moved a question into the focus of the public eye that has occupied philosophers for decades: Under what conditions is it appropriate to assert a certain claim? When asserting a claim that x, must one know that x? Must x be true? Might it be normatively acceptable to assert whatever one believes? In the largest cross-cultural study to date (total n = 1,091) on the topic, findings from the United States, Germany, and Japan suggest that, in order to claim that x, x need not be known, and it can be false. However, the data show, we do expect considerable epistemic responsibility on the speaker’s behalf: In order to appropriately assert a claim, the speaker must have good reasons to believe it.


Results
Factive accounts predict assertability to be low in the false belief condition -at the very least significantly below chance. However, in line with predictions of nonfactivist accounts of assertion, at least 3 out of 4 participants considered it perfectly acceptable to assert a false claim in all three countries (significantly above chance, binomial tests, all ps<.001, two-tailed), see Table 1. The results demonstrate quite clearly that the norm of assertion is not factive. Expectedly, in the true belief condition, assertability exceeded 97% in all three countries. Both manipulation checks suggest that the vignettes worked well: In the true belief condition, nearly everyone considered the proposition at stake true (at least 93%, significantly above chance, binomial tests, all ps<.001), and hardly anybody considered it true in the false belief condition (at most 9%, significantly below chance, binomial tests, all ps<.001). In both conditions justification was perceived to be high (at least 68%, significantly above chance, binomial tests, all ps<.003), although -in both conditions -it was a little lower for Japanese subjects in comparison to US Americans and Germans.
Due to a cell count of zero for "unassertable" in the German true belief condition (which is consistent with the hypotheses of all accounts), a binomial logistic regression analysis could not be performed. Aggregating across conditions, a Pearson Chi Square test revealed no significant difference in assertability judgments across countries (χ 2 (2,n=461)=.56, p=.755, f=.035).
Aggregating across countries, truth did have a significant, though not particularly large, impact on assertability (χ 2 (1,n=461)=43.05, p<.001, f=.31).  Including the excluded participants (less than 10%) in the sample left the results by and large the same (see Table 2). It did, however, make it possible to conduct a regression analysis (we no longer have a cell count of zero for "unassertable" in Germany), see Table 3. Truth had a significant impact on assertability (p=.007, Odds Ratio=.123), Country and the interactions were nonsignificant (all ps>.227). Although truth was significant, note that the model explained less than 20% of the variance (Nagelkerke R 2 =.196).

Study 2 2.1 Participants
A total of 596 participants were recruited on online platforms in the US (Amazon Mechanical Turk), Germany (Clickworker) and Japan (Crowdworks). As preregistered, 3 participants whose native language was not English, German or Japanese respectively, and those who failed an attention or comprehension check were excluded, leaving 575 participants (310 female, age M=41 years, SD=12 years). The subsamples for each country were similar in terms of gender distribution and average age. 4

Materials 2.2.1 Materials in English
The experiment, based on Kneer (2018), took a 2 justification (good v. bad evidence) × 2 question formulation ("should have said" v. "appropriate to say") between-subjects design. Participants were randomly assigned to one of the four conditions. In the scenario, justification was manipulated, varying between good evidence and bad evidence (in square brackets): Carlos is at the JFK airport in New York, waiting for his flight to Amsterdam. An elderly woman asks him whether he could tell her at which gate the flight to Paris departs. Carlos has a look at the monitor listing the departure gates. [The departure list states that the only flight to Paris leaves at gate 24. / He cannot find the flight to Paris in the list. But Carlos has a hunch that it might depart from gate 24, and thus comes to believe that it will.] Carlos says to the elderly woman: "The flight to Paris leaves at gate 24." The assertability question came in two variations ("should have said" v. "appropriate to say"

Results
Due to a cell count of zero in several conditions (see Table 5, all consistent with predictions), it was impossible to run a binomial regression analysis with all three factors, i.e. formulation, justification and country. I thus conducted Pearson's Chi Square test, which revealed no significant difference in assertability judgments across formulations (χ 2 (1,n=575)=.52, p=.469, f=.030). Averaging across formulations makes it possible to conduct a regression analysis with the two key predictors: justification and country. The analysis revealed a significant and powerful effect for justification (p<.001, Odds Ratio=913.75), see Table 6. Unsurprisingly, the proportions of participants who considered the proposition at stake assertable yet judged it unjustified was significantly below chance in any condition (at most 11%, binomial tests, test proportion =.25, 5 all ps<.001, one-tailed). Overall country, as well as all interactions, proved nonsignificant (all ps>.100). The model explained more than 80% of the variance (Nagelkerke R 2 =.838). As a manipulation check, participants were consulted as to whether they held that the agent actually believed that p in the first place. In any condition in any country at least about four in five participants responded with yes (significantly above chance, binomial tests, all ps<.001).