Indirect measures (Limitations of IAT (Behaviour:
Oswald et al. (2013)…
Limitations of IAT
- Oswald et al. (2013) only weakly predicted behaviour, and was no better than explicit measures
- BUT, argued to depend on the concepts included: Greenwald et al. (2009) IAT tended to better predict socially sensitive concepts e.g. discrimination, mental health, than explicit measures. But for less socially sensitive concepts e.g. political preferences, explicit/direct better predict
= may be better depending on types of concepts --> not particularly problematic considering unconscious attitudes often researched involve socially sensitive attitudes that p's may not want to consciously disclose
Fielder et al. (2006) proposed many limitations:
- assumes that slow reaction times to paired primes have a causal relationship with attitudes, yet this cannot be inferred--> target words used tend to be ambiguous (e.g. ‘flower’ for positive stimuli and ‘insect’ for negative stimuli), so might not reflect attitudes towards the concepts intended, such as race
- assumes that valence is representing the same attitude e.g. slow reaction to black-positive stimuli pair, and a fast reaction to black-negative stimuli pair represent prejudice towards black people. BUT Stephens and Buschner (2008) found a significant difference in the reliability measure for each positive and negative valence task = they should not be considered to be measuring the same attitude.
- IAT quite obvious in what it's testing = not actually testing unconscious/automatic thoughts. Plausible cos increasing awareness about the test can change outcomes --> Wallaert et al. (2010) prompt from experimenter to avoid stereotyping sig. reduced pro-White bias, explicit instruction to stereotype sig. increased pro-White bias = susceptible to external control?
BUT quite difficult to intentionally fake responses- Egloff and Schmuckle (2002) p's failed to portray themselves positively when explicitly asked to do so on IAT, but could on anxiety self-report measure
Means reaction times might be influenced by other factors e.g. resource depletion, cognitive fluency
- McFarland and Crouch (2002) slower overall response times lead to more extreme IAT scores. Older p's also tended to have higher IAT scores = could be related to them having slower overall response times.
= Original IAT may therefore not be sensitive enough to cognitive fluency, but updated version by Greenwald et al. (2003) provides a new algorithm that controls for this confound BY.....
- Cultural knowledge: Arkes & Tetlock (2004) results of the IAT reflect shared cultural stereotypes rather than personally held ones .
BUT, even if arise from culture, they are still implicit and are attitudes that can influence behaviours (Banaji et al. 2004)
IAT actually reflects differing prior exposure to the concepts, with positive bias towards more familiar stimuli (Ottaway et al. 2001)
- Zajonc (1968) having had less interaction with the ethnic group being studied might be associated with slower reactions on incongruent trials because associating positive stimuli with less familiar items is more difficult
- Ottaway et al. (2001) investigated the role of familiarity and frequency using an IAT looking at implicit racism towards both African American and Hispanic groups. Even when name familiarity and frequency were controlled for, p's held implicit racism towards these groups. IAT was more sensitive when high familiarity names were used, yet low familiarity names didn’t eliminate its sensitivity = familiarity does not account for race effects
- Single category IAT (Karpinski & Steinman, 2006) makes it possible to investigate the association for a single target concept or a single attribute, rather than comparing presence for one large over another --> bypasses issues of valence and has had success to date
- Personalised IAT (Olson & Fazio, 2004) controls for external influences on the traditional IAT --> uses category labels less likely to be influenced by normative information e.g. "I like" and "I don't like". Aims to allow p's to evaluate the concepts and has been associated with reduced racial prejudice and correlates more strongly than the original IAT with explicit attitudes
Extrinsic Affective Simon Task (De Houwer, 2003)
- p's categorise stimuli consisting of words that either have positive or negative valence presented in either the color white or two different colors.
- When words are presented in white, p's categorise words on their perceived positive or negative valence (e.g. nice = +ve, boring = -ve). When the words are presented in color, p's are asked to categorize based on color alone and ignore word meaning.
- When colored words are presented, categorization accuracy and speed are facilitated when, for words which the respondent has a positive implicit attitude, the response was the same as was expected for white words with obvious positive valence.
- e.g. EAST designed to measure evaluations of alcoholic beverages, p's may be presented with positive and negative words in white ink (e.g., spider, sunrise) and with names of alcoholic and non-alcoholic beverages (e.g., beer, soda) that are presented in yellow ink on some trials and in blue ink on others.
- P's task is to press a left-hand key when they see a white word of negative valence or a word printed in blue ink and to press a right-hand key when they see a white word of positive valence or a word printed in yellow ink.
- To the extent that participants show faster (or more accurate) responses to a colored word (e.g., beer) when the required response to this word is combined with a positive as compared to a negative response, it is inferred that participants showed a positive response to the object depicted by the colored word.
- Typical EAST includes 3 blocks, 2 practice and 1 critical.
- allows to test single categories (IAT doesn't)
- Can allow to assess single associations and multiple associations
- EAST effect can be calculated by comparing trials within the same task
- EAST effects are less likely to be influenced by non-associative variables that determine how participants recode tasks
- Measures evaluative responses and self-related associations --> BUT, not as reliable: Teige et al. (2004) EAST was adapted to allow the simultaneous assessment of the three traits shyness, anxiousness, and angriness. In order to test the EAST’s psychometric properties, 100 participants completed a trait EAST, Implicit Association Tests (IATs), and direct self-ratings. The EAST showed low internal consistencies and correlated neither with the IATs nor with the direct measures
- Use e.g. De Jong (2007) looked at attitudes of heavy and light drinkers towards alcohol and soda. IAT has previously shown +ve associations with soda, which seems counterintuitive --> Heavy and light drinkers showed positive associations with sodas and neutral automatic associations with alcohol (argue that +ve towards soda doesn't represent -ve towards alcohol, but neutral attitude). Importantly, positive automatic alcohol associations predicted variance of alcohol (mis)use and was the single best predictor of individuals' alcohol problems = implicit attitudes towards alcohol play a role in alcohol (mis)use.
- Lower reliability than other tasks
- De Houwer and De Bruycker (2007) low reliability of EAST due to the fact that p's do not have to process the meaning of the colored target stimuli for the color-based responses in the task. To overcome this, they developed a modified version of the EAST in which p's are forced to process the meaning of the target stimuli.
- The identification EAST (ID-EAST) includes presentations of target and attribute words in upper and lower cases. Positive and negative attribute words have to be categorized in terms of their valence irrespective of whether they are displayed in upper or lower cases; the target words have to be categorized depending on whether they are presented in upper or lower cases.
e.g. in an ID-EAST designed to measure evaluative responses to beer, p's may be presented with positive and negative words and the word beer in either upper or lower cases. Participants’ task would be to categorize the attribute words in terms of their valence by pressing one of two response keys. However, for the word beer, participants would be instructed to press one response key when it is presented in upper cases and the opposite key when it is presented in lower cases. Because the attribute words are also presented in upper and lower cases, participants are therefore required to process the semantic meaning of the word beer before they are able to identify the correct response key.
- This procedural modification increased the reliability of the EAST, although it is still somewhat lower than the average reliabilities of the IAT
Intro & Conclusion
- Argued that people are aware of what is being assessed by an explicit/direct measure (bias, lie) but unaware of what is being assessed by an implicit/indirect measure (Petty, Fazio & Brinol, 2009)
- Explicit/direct = self-report scales, psychometric tests
- Implicit/indirect = semantic priming tasks, GoNoGo association task
- field would benefit from a stronger focus on underlying mechanisms with regard to the measures themselves as well as their capability to predict behaviour
--> The groundwork for this focus has already been set by the development of mathematical modeling techniques (e.g., Conrey et al., 2005; Klauer et al., 2007; Payne, 2008), in which measurement outcomes are treated as behaviors that are themselves in need of a psychological explanation rather than as direct reflections of mental constructs (e.g., automatic associations) that can be used to explain behavior.