Overall Discussion

The experiments presented in this chapter provide evidence of a hitherto overlooked role of test question format as a source of goal information in memory evaluation. The behavioural trends of Experiment 1 suggested that test questions adopted as standard in recognition research are capable of influencing memory decision outcomes. Experiment 2 followed up these preliminary findings with systematic manipulation of memory strength, and recovered heightened test question effects under conditions of low overall memory strength. Experiment 3 tested two competing explanations for the question effects, and found evidence in favour of a goal

emphasis mechanism rather than a habitual preference for responding “no”. In all three experiments, we replicated the effect of question emphasis shifting SDT

criterion (as estimated from two prominent variants of the recognition SDT model) to instil a disconfirmatory bias that reduced the likelihood of endorsing the emphasised decision. Our analyses involving the more unorthodox decision accuracy measures also recovered a consistent effect, wherein question emphasis improved the

accuracy of endorsements of the emphasised decision. Considered with the criterion bias effects, the pattern of results suggests that question emphasis improves the accuracy of making emphasised responses at the expense of their frequency, consistent with the controlled engagement of strategic caution. These results are now discussed with reference to previous findings in the fields of memory and decision-making.

Prior studies investigating effects of test format in standard single item recognition have typically focused on measures of source memory or subjective experiences of memory (Bastin & Van der Linden, 2003; Dodson & Johnson, 1993; Hicks & Marsh,

102 1999; Khoe, Kroll, Yonelinas, Dobbins, & Knight, 2000; Marsh & Hicks, 1998). For example, Hicks and Marsh (1999) reported an increase in “old” responding in a recognition test phase comprising three response options with two “old”

subcategories (“remember old”/“know old”/“new”) compared to a test phase with two

equally weighted response options (“old”/“new”, followed by “remember”/“know”;

though cf. Bruno & Rutherford, 2010). Here we present evidence of a test format bias in the primary measure of recognition performance, which manifests in the absence of unequally weighted response options. Nevertheless, our findings accord with previously reported test format biases in highlighting the salient influence of aspects of the test environment on the outcomes of memory evaluation (Johnson et al., 1993).

Further, the observed disconfirmatory bias was in the opposite direction to previously reported test format biases, such as the acquiescence bias in questionnaire

responding (‘yea-responding’; Podsakoff, et al., 2003), the Loftus framing effect in eyewitness testimony (Loftus, 1996) and decision cueing in recognition research (e.g. O'Connor, et al., 2010; Rhodes & Jacoby, 2007). We suggest that our

emphasis-driven bias differs from these other biases in that it establishes a decision goal against which diagnostic evidence is evaluated, rather than contributing to the evidence itself. This aspect serves to distinguish the present question effects from the classic Loftus framing effect, in which questions that framed details of potentially encoded contexts (“leading questions”) led to a confirmatory response tendency that enhanced fallacious endorsement of contexts as having been previously

experienced (Loftus, 1996). Our questions lacked any such allusion to explicit aspects of encoding, which would putatively impact upon memory evaluation by directly interfering with the assessment of memory evidence i.e. via the retrieval

103 control processes outlined in the previous chapter (see Section 1.3.). The question bias is similarly differentiated from the cueing bias, which despite also acting on the same broad category of decision control processes (operating independent of memory strength) as the emphasis bias, nevertheless serves to instil cued

expectations that serve as an additional diagnostic evidence base for responding. Rather, our question manipulations impacted via a different strategy, by implicitly emphasising a particular memory decision as a higher order goal, the endorsement of which demanded a higher overall level of diagnostic evidence relative to the non- emphasised decision. Our findings therefore inform a tentative distinction between three major strategic biases wrought by different aspects of the retrieval

environment: decision cueing (confirmatory), evidence framing (confirmatory) and goal emphasis (disconfirmatory). Chapter 4 examines the interaction of cueing and goal emphasis manipulations with a view to directly interrogating the above bias trichotomy.

The present effects hence reveal a strategic influence on memory evaluation that is potentially consistent with findings in the rewarded decision-making literature,

particularly mixed incentive research. In mixed incentive research, monetising one of two available response options leads to a reluctance to endorse the monetised decision (Newman et al., 1985). A similar behavioural trend indicative of this disconfirmatory bias was reported in a previous recognition experiment that differentially provided mixed incentives for “old” or “new”decisions (Han, Huettel, Raposo, Adcock & Dobbins 2010; see also Section 1.5.). The present findings are notable for inducing comparable caution without the provision of incentives, and demonstrate that even in the absence of any explicit mention of reward, participants display a goal-directed bias that is driven by implicit information gleaned from the

104 testing environment. Chapter 4 directly tests the correspondence of the present question emphasis effects with more explicit emphasis mechanisms engaged by mixed incentives.

We also found evidence that the test question bias, whilst decreasing the frequency of the emphasised decision, improved the accuracy of these decisions when they were made. As outlined in the previous chapter, manipulations enacted solely at retrieval have little or no impact on old/new discrimination sensitivity, which is largely determined by encoding processes (Craik & Lockhart, 1972; Stretch & Wixted, 1998; albeit with some exceptions; Dobbins & McCarthy, 2008; Marsh & Hicks, 1998). Indeed, across all three experiments our retrieval manipulations had little effect on measures of d′ sensitivity that binned responses by ‘objective’ item categories and collapsed across old and new decisions. However, the observed effects on decision accuracy measures oldcorr and newcorr (which binned responses by ‘subjective’ decision categories) highlight the efficacy of these more unorthodox analyses in elucidating a strategic effect which would otherwise be somewhat obscured by a reliance on SDT analyses alone. Analogous influences of the test environment on strategic use of retrieved memory evidence have been highlighted previously by models of memory control (Koriat & Goldsmith, 1996). In the present context, it furthers our suggestion of a goal-directed mechanism underlying the question format effects, which prompted participants to adjust their criterion towards a more rigorous evaluation of diagnostic evidence promoting the emphasised decision. The question therefore served to instil the goal of getting a greater proportion of the emphasised decisions correct, even if that entailed making fewer endorsements of that decision. Overall, the present findings allude to an implicit method by which the test

105 yet neglected environmental precipitant of memory control identified in the previous chapter (see Section 1.1. and 1.2.). This provides preliminary evidence for the previously speculated propensity for participants in a memory test to infer higher- order goals from uncontrolled environmental aspects (Johnson et al., 1993; Koriat & Goldsmith, 1996). Further, the results also validate SDT criterion as a behavioural index of memory control processes, albeit whilst also demonstrating the virtue of conducting complementary analyses of other behavioural measures to enable stronger inferences (indeed, this complementary analytic technique will be

maintained in subsequent chapters). The real-world implications of these laboratory findings are made apparent by revisiting the previous eyewitness test scenario in which heightened goal emphasis acted to reduce the likelihood of identifying a suspect in a lineup as the perpetrator of a crime. The present findings raise the additional possibility that cross-examiners might be capable of biasing eyewitness reports merely by selectively mentioning either “old” or “new” decisions in their manner of questioning (especially if the strength of the eyewitness’ memory of the crime itself is weak). The real-world significance of these goal emphasis effects inspired further investigation in ensuing experimental chapters.

The most immediate avenue of investigation followed from the literature review of the previous chapter, which highlighted a neglect in determining whether

environmentally provoked strategies observed in episodic memory evaluation also manifest in other cognitive domains (see Section 1.4. and Section 1.6.). This represents an important step in elucidating the cognitive underpinnings of behaviourally observable memory strategies, and specifically whether such

strategies are rooted in control processes confined to the episodic domain (“domain- specific” control) or those that are capable of modulating behaviour in a number of

106 different domains (“cross-domain” control). Hence, the next experimental chapter investigated the emergence of the presently established question emphasis bias across episodic and non-episodic domains.

107

Chapter 3: Question format as an implicit source of goal emphasis