«Philosophy and Phenomenological Research Philosophy and Phenomenological Research Vol. XC No. 1, January 2015 doi: 10.1111/phpr.12160 © 2015 ...»
Philosophy and Phenomenological Research
Vol. XC No. 1, January 2015
© 2015 Philosophy and Phenomenological Research, LLC
Assertion and Assurance: Some
University of Waterloo
I report three experiments relevant to evaluating Krista Lawlor’s theory of assurance,
respond to her criticism of the knowledge account of assertion, and propose an alternative
theory of assurance.
Introduction Krista Lawlor’s Assurance is a rich and stimulating work. Lawlor argues that assurance is “a distinctive speech act,” that assurance “is not the speech act of assertion,” and that assurance has a special relationship to claiming that one has knowledge (Lawlor 2013, quotes from pp. 9, 3; unless otherwise noted, all parenthetical page references are to this work). Because she theorizes about the nature and function of some familiar aspects of ordinary language, her theory is answerable to empirical facts about actual usage.
Lawlor admirably acknowledges this constraint. She bases her theory on “facts about usage” (p. 22), which constitute “data that any account” in this area “must... respect and explain” (p. 80).
I do three things in this paper. I report three simple experiments that test whether Lawlor’s theory is headed in the right direction. I evaluate what Lawlor describes as “compelling criticisms of the knowledge account of assertion.” And I suggest an alternative theory of assurance.
Recognizing a Potential Change of Mind: Experiment 1 Lawlor looks to the “normative dimensions of assurance” to distinguish it from assertion (pp. 11–15). Assertion and assurance, Lawlor hypothesizes, differ in how one represents oneself when performing them. When one asserts that a proposition is true, “it is recognized that there may be further 214 JOHN TURRI evidence or circumstances that change [one’s] opinion” on the matter. By contrast, when one assures that a proposition is true, it is not recognized that further evidence or circumstances might change one’s opinion. Instead, one represents oneself as having “conclusive reasons” that show any further counterevidence to be misleading. If Lawlor’s hypothesis is true, then it should have detectable consequences for how people think conversational partners should interpret one another.
To test this hypothesized difference, I conducted a simple experiment with eighty participants.1 Participants were randomly assigned to one of two conditions, Assertion and Assurance. Each participant read a single story about Jeff and Sally. The only difference between the two stories is that in the Assertion story Jeff makes an unadorned assertion using a simple declarative sentence, whereas in the Assurance story he explicitly adds “I assure
you.”Here is the text:
Jeff and Sally are driving to their appointment. They left right on schedule, but they had to take a detour because of some road construction. Sally asks if they will arrive by ﬁve o’clock. Jeff replies, “We will arrive by ﬁve o’clock[, I assure you].” After reading the story, participants rated their agreement with the key test
1. Sally should recognize that, depending on how things go, Jeff might change his mind about when they’ll arrive.
Responses were collected on a 6-point Likert scale anchored with “Strongly Disagree” (-3), “Disagree” (-2), “Somewhat Disagree” (-1), “Somewhat Agree” (1), “Agree” (2), and “Strongly Agree” (3), appearing left-to-right across the participant’s screen. (Participants never saw the numerical labels, only the qualitative anchors. The same is true in all experiments below.) The ﬁrst critical question is whether participants in the Assurance condition were more likely to disagree with the test statement. If they were, then it supports Lawlor’s hypothesis. But if they were not, then it undermines Lawlor’s hypothesis. They were not. Mean response in the Assurance condition (M = 0.70, SD = 1.79) did not differ from mean response in the Assertion condition (M = 0.93, SD = 1.69).2 A second critical question is
Aged 18–67, mean age = 33; 35 female; 92% native English speakers. Participants were
recruited and tested online (using Amazon Mechanical Turk and Qualtrics) and compensated $0.35 for approximately 2 minutes of their time. Repeat participation was prevented. For all subsequent experiments reported here, I recruited and compensated all participants similarly.
Independent samples t-test, t(78) = 0.58, p =.564. All reported tests are two-tailed.
BOOK SYMPOSIUMwhether participants in the Assurance condition tended to disagree with the test statement. If Lawlor’s hypothesis is true, then we should expect them to disagree. But this is not what we observe. Instead participants in each condition tended to agree.3 The mode response was “Agree” in both Assertion (40%) and Assurance (45%) conditions.
To ensure that the manipulation was effective, I included three additional probes. Each appeared on a new screen after the key test statement. (Participants could not go back and change answers, and the story remained at the top of the screen throughout.) The second and third probes asked participants to rate their agreement, on the same 6-point scale as above, with the
2. Jeff assured Sally that they will arrive by ﬁve o’clock.
3. Jeff told Sally that they will arrive by ﬁve o’clock.
As expected, mean agreement with the second probe was higher in the Assurance condition (M = 2.48, SD = 0.88) than in the Assertion condition (M = 2.08, SD = 0.86).4 Mean agreement with the third probe did not differ between the Assurance (M = 2.43, SD = 0.98) and Assertion conditions (M = 2.43, SD = 1.15).5 For the ﬁnal probe, participants were asked, “Which better describes what Jeff did?” and then presented with an open sentence.
4. He _____ Sally that they will arrive by by ﬁve o’clock. (told/ assured) Responses options for the fourth probe were rotated randomly. As expected, signiﬁcantly more participants in the Assurance condition (93%) than in the Assertion condition (50%) answered “assured.”6 Participants in the Assurance condition chose this answer at rates far exceeding chance,7 whereas participants in the Assertion condition were evenly split.8 These One sample t-tests, test proportion = 0. Assertion, t(39) = 3.47, p =.001, MD = 0.93,
Chi-square test for goodness of ﬁt, v2(n = 40, df = 1) = 0.00, p = 1. Within the Assert condition, response to the ﬁnal probe did not predict response to the key test question, linear regression, Beta =.075, p =.645.
216 JOHN TURRI results show that the manipulation was effective and participants are sensitive to the difference between assuring and (merely) asserting.
Lawlor hypothesized that we represent ourselves as having very different evidence when we make an assurance than when we make a mere assertion. This is supposed to be reﬂected in what we “recognize” for assurances and assertions. More speciﬁcally, we recognize that an asserter might subsequently change his mind in light of new developments, whereas we do not recognize that an assurer might do this. The results from the present experiment suggest that this is not reﬂected in how we think people should interpret one another. However, Lawlor considers this “just a ﬁrst rough pass in our effort to understand” what distinguishes assurance from assertion. The next section looks at another hypothesis she offers.
Natural Conjunctions: Experiment 2 Lawlor hypothesizes that “the function of assurance is to give hearers exclusionary reasons,” whereas this is not the function of assertion (pp. 15–22).
An exclusionary reason to believe a proposition allows one to “disregard other reasons one might have against” the proposition, which enables us to properly “stop weighing reasons and get on about our business” (pp. 18, 20). Lawlor argues that “usage supports” this hypothesis. In particular, she claims, “It is natural to say ‘p, but see for yourself’; it’s not so natural to say ‘I know p, but see for yourself’” (p. 19). (She assumes that saying “I know” a proposition counts as an assurance that the proposition is true.) To test the claim about what is natural, I conducted a simple experiment with eighty-two new participants.9 Participants were randomly assigned to one of two conditions, Assertion and Assurance. Each participant read a single story about Jeff and Sally, this time discussing what to eat for dinner.
The only difference between the two stories is that in the Assertion story Sally makes an unadorned assertion using a simple declarative sentence, whereas in the Assurance story she adds “I know.” In each story Sally ends by adding the independent clause “but go ahead and see for yourself.” Here
is the text:
Jeff and Sally are deciding what to eat for dinner. Sally looks in the refrigerator and notices that they have leftovers. Jeff asks whether the leftovers have been sitting around for too long. Sally answers, “[They are/I know they’re] ﬁne, but go ahead and see for yourself.” Aged 18-59, mean age = 29; 24 female; 95% native English speakers.
Responses were collected on a 6-point Likert scale anchored with “very unnatural” (-3), “unnatural” (-2), “somewhat unnatural” (-1), “somewhat natural” (1), “natural” (2), and “very natural” (3), left-to-right across the participant’s screen.
The critical question is whether participants in the Assurance condition were more likely to view Sally’s answer as unnatural. If they were, then it supports Lawlor’s hypothesis. But if they were not, then it undermines Lawlor’s hypothesis. They were not. Mean response in the Assurance condition (M = 1.10, SD = 1.43) did not differ from mean response in the Assertion condition (M = 1.00, SD = 1.58). Participants in the Assurance condition tended to rate the statement as natural,10 and no less natural than in the Assertion condition.11 The mode response was “natural” in both Assertion (44%) and Assurance (44%) conditions.
As a check on the robustness of the ﬁndings from Experiment 1, I included a second probe on a new screen.
2. Jeff should recognize that, depending on how things go, Sally might change her mind about whether the leftovers are ﬁne.
Responses were collected on the same 6-point Likert scale used for the analogous question in Experiment 1. The results replicated the earlier ﬁndings. There was no difference in mean agreement between the two conditions,12 and overall the trend was for people to agree that Jeff should recognize that Sally might change her mind depending on things go.13 Lawlor hypothesized that assurance, but not assertion, functions to provide others with reasons that allow them to discount potential counterevidence, end deliberation and act. She supported this hypothesis by claiming that it reﬂects ordinary usage. In particular, she said that although it is natural to follow up an assertion with “but see for yourself,” it is not natural to follow up an assurance that way. The results from the present experiment One sample t-test, test proportion = 0, t(40) = 4.92, p.001, MD = 1.10, 95% CI =
218 JOHN TURRI suggest that this is false. Adding “but see for yourself” is natural for both assertion and assurance.
Eavesdropping and Blame: Experiment 3 Thus far we’ve evaluated two different hypotheses for distinguishing assurance from assertion. Neither seems to work. After presenting the hypotheses, Lawlor proposes to focus more directly on understanding “the normative demands of assurance giving” (p. 21). Again she appeals to “facts about usage”—and in particular “one revealing fact about usage”—to advance her inquiry. The alleged fact is this: if one person offers assurance to a second party, then a third party eavesdropping on the exchange can treat the ﬁrst person “as a guarantor” and “blame her if she is in error.” Lawlor calls this a “datum” that shows us “something important about the extent of one’s commitment in giving an assurance.” In particular, she thinks it shows that someone who offers an assurance is thereby accountable to “all hearers, actual or potential” (p. 22).
To test this whether this is a fact, I again conducted a simple experiment with eighty-three new participants.14 Participants were randomly assigned to one of two conditions, Partner and Eavesdropper. Each participant read a single story about Jeff and Sally, this time discussing investment in biotechnology. In each story Sally assures Jeff that a certain ﬁrm is the best investment, and an eavesdropper, Penelope, overhears this. The only difference between the two stories is that in the Partner story Jeff invests in the ﬁrm, whereas in the Eavesdropper story Penelope invests in the ﬁrm. In each
story the investment is a disaster. Here is the text:
Jeff and Sally are having a private conversation about which biotechnology ﬁrm is the best investment over the next year. Sally says to Jeff, “NanoGenes is the best investment, I assure you.” It turns out that another person, Penelope, is in the next room eavesdropping on their conversation.15 Based on what Sally said, [Jeff/Penelope] invested in NanoGenes. It was a disaster and [Jeff/Penelope] lost all the money [he/she] invested.
Participants then responded to the key test statement:
1. [Jeff/Penelope] _____ blame Sally for being wrong about NanoGenes.
Responses were collected on a 6-point Likert scale anchored with “deﬁnitely cannot” (-3), “cannot” (-2), “probably cannot” (-1), “probably can” (1), Aged 19–66, mean age = 35; 44 female; 92% native English speakers.
Indicates paragraph break on the participant’s screen.