Indirect attitude measures evaluate implicit attitudes – they assess unconscious mental associations between different concepts, using participants’ performance on experimental paradigms, most often speed categorisation tests. They were created to overcome the shortcomings of direct measures, which are a way of assessing attitudes directly – that is, a participant reports their own attitudes consciously, for example through a self-report questionnaire. There are many different implicit attitude measures, and the purpose of this essay is to critically evaluate their usefulness.
Indirect attitude measures, such as the Implicit Associations Test, the Evaluative Priming Task and the Affect Misattribution Procedure were designed to overcome the problems with direct measures. There are many complications with using self-report methods to assess attitudes, all of which impact the reliability and validity of the results. For example, Paulhaus (1991) pointed out that self-reports are subject to social desirability and self-presentational concerns, as people can control their answers – this means they could give socially desirable responses to avoid trouble or embarrassment, causing the replicability of the results to be compromised. Direct measures are also susceptible to demand characteristics because the theme of the questions could give away the hypothesis, and participants could respond with what they think the researchers want to hear, rendering the results unreliable. Wilson and Dunn (2004) noted the limited ability of people to report their attitudes, as they may be unaware of their unconscious biases and associations. This would further hinder the results gathered as their reported attitudes may not reflect their true attitudes, impacting the validity of the measure. Schwartz (1999) stated that direct measures are notoriously sensitive to contextual factors that are unrelated to the attitude itself, for example who is asking the questions, what time of day it is, the temperature of the room and more. All of these factors could influence the results given by the participant. As you can see, direct measures have a lot of issues to overcome, all of which must be considered when using these measures.
The Implicit Associations Test (IAT) created by Greenwald, McGhee and Schwartz (1998) measures reaction times in compatible versus incompatible blocks. The underlying idea is that you should be quicker to categorise words when closely related items share the same response key, i.e. a faster performance in compatible blocks suggests that the evaluation made by the participant is consistent with the stereotype being measured. A standard IAT consists of 5 blocks – the first 2 blocks are practice blocks, where, if measuring racism for example, the participant would categorise the images into black or white faces by pressing certain keys. This is the initial target-concept discrimination. They would then categorise words into good and bad words, which is the initial attribute discrimination. In the third block, the first two practice tasks are combined with the same key assignments, and this is the initial combined task. The fourth block is almost the same as the first, but the key assignment is reversed. This is the reversed target-concept discrimination task. The fifth and final block combines the two categorisation tasks, but the key assignments from the first and second block are reversed – aptly named the reversed combined task. The general finding is that participants respond slower to incompatible blocks. Quick and accurate responses are facilitated when the key assignments combine concepts that are strongly associated in memory.
There are many advantages of using the IAT to assess implicit attitudes. The IAT’s popularity is partly due to its satisfactory psychometric properties (Tiege-Mocigemba et al. 2010), so much so that the IAT’s reliability estimates are comparable to self-report scores (Gawronski and De Houwer, 2014), meaning that results gathered from either can be compared with each other. The flexibility of the IAT is to be admired – it can be used to compare the relation between any two concepts, making it a very useful indirect measure. Variants of the IAT also make it possible to assess associations of a single concept rather than relative associations of two concepts (Karpinski and Steinman, 2006) which means that it has an even wider applicability, making it more likely to be a psychologists’ first choice. It also has high reliability (ranging from .70-.90; Gawronski and De Houwer, 2014), making it an optimum choice for assessing implicit attitudes.
However, there are also a number of limitations with using the IAT. For example, it has been found that participants can be instructed to fake their scores (e.g. De Houwer, Beckers and Moors, 2007), suggesting that this measure is susceptible to manipulation, making it less reliable than it first appears. The metric that the IAT is scored on may also be problematic (Blanton and Jaccard, 2006), further decreasing the reliability of the results. The presentation of compatible and incompatible blocks in separate, consecutive blocks could distort scores through different sources of systematic error variance (Tiege-Mocigemba et al. 2010), causing the results to be invalid. The IAT can also be influenced by factors unrelated to the to-be-measured concepts, like the block structure, the order of compatible and incompatible blocks, and reordering processes (Tiege-Mocigemba et al. 2010). Reordering processes occur when participants try to recode the four categories into two using their own parameters; this lowers validity as the parameters that each participant uses will be subjective and different for each participant. Finally, this measure is comparative in nature: it assesses relative evaluations for one concept to another but does not give individual scores for each group, damaging the suitability for association measurement to an individual target.
There have been some adaptions to the IAT, trying to compensate for its limitations. For example, the Single Category IAT (SC-IAT; Karpinski and Steinman, 2006) attempted to cope with the comparative nature of the IAT by reducing it to a single category. This measure has demonstrated its value in a considerable number of studies. The Recoding-Free IAT (RF-IAT; Rothermund et al., 2009) and the Single Block IAT (SB-IAT) have replaced the original block structure, instead mapping response keys across trials. This adaption has the advantage of being uninfluenced by order effects of compatible and incompatible blocks (Nosek, Greenwald and Banaji, 2007). The Single Target IAT (ST-IAT; Bluemke and Friese, 2008) improves the IAT by allowing for the measurement of associations between evaluations and single categories (e.g. black/good and black/bad). These adaptions have all shown vast improvement on the standard IAT and support the usefulness of this measure.
Another implicit attitude measure is the Evaluative Priming (EP) Task, by Fazio, Sanbonmatsu, Powell and Kardes (1986). This measure employs the basic procedure of sequential priming to assess evaluative responses (Fazio et al., 1986), and is based on the assumption that the activation of one concept automatically activates related concepts. To the extent that an attitude is strong, the activation of one concept automatically activates an evaluation; responding to a target stimulus should be faster to the extent that the prime and evaluative target are associated. A typical EP trial, in the example of racism, consists of briefly seeing a picture of a black or white person (the prime; e.g. 200ms) and then a blank screen for a short time (e.g. 50ms), and then a positive or negative word (the target). The objective is to press one key (e.g. S) if the target is positive, and another (e.g. K) if the word is negative – this is the evaluative decision task. There would be a stronger association between black and bad relative to white and good inferred if pictures of black people facilitate responding to negative words more than picture of white people facilitate responding to negative words. A number of variants of this implicit measure exist, but it’s recommended to instantiate the measure by presenting three blocks of 60 trials each (Fazio, Jackson, Dunton and Williams, 1995).
There are a number of reasons why a researcher might use the EP task. For example, the mechanisms and outcomes of the EP task have been studied more thoroughly than nearly any other measure, giving researchers detailed guidance in its use (e.g. De Houwer, Tiege-Mocigemba, Spruyt and Moors, 2009), meaning that researchers know how to get the most reliable results, making this a key measure. A major benefit of the EP is that researchers can calculate associations between different primes and target categories separately, as the use of a neutral baseline allows you to separately estimate associations between (in the context of racism) black and good, black and bad, white and good, and white and bad. Therefore, it allows researchers to calculate separate priming scores for different kinds of associations that are confounded in the IAT (Wittenbrink, 2007), making this measure more useful than the IAT as it gives more information.
However, there are also limitations to consider. Like the IAT and the SC-IAT, it’s possible that a motivated participant could deliberately control the outcome of the task (Tiege-Mocigemba and Klauer, 2008), suggesting that this measure is also susceptible to manipulation, and controls must therefore be put in place. The reliability of this measure is notoriously low, rarely exceeding .50 and commonly even lower than that (Gawronski and De Houwer, in press) – this is a very low reliability score, suggesting that this measure is not the most appropriate to use, as the reliability of the results is questionable.
Another implicit measure is the Affect Misattribution Procedure (AMP) by Payne, Cheng, Govorun and Stewart (2005), which measures automatically activated responses based on misattributions participants make about the sources of their affect. This measure is based on the idea that valence of a perceived stimuli will automatically transfer to a subsequently-presented neutral stimulus. A standard AMP consists of trials with a prime, followed by a blank screen, followed by a Chinese ideograph, followed by a black and white mask. The task is to indicate whether they consider the Chinese ideograph visually more or less pleasant than the average Chinese character. The general finding in an AMP is that neutral Chinese ideographs are evaluated more positively when participants are primed with a positive rather than a negative stimulus. Payne et al. (2005) assumed that the underlying mechanism is affect misattribution, where the actual source of the affect – the prime – is incorrectly identified, and the affect that is triggered by the prime is mistakenly attributed to the target. The AMP is not scored by comparing response latencies like other implicit measures but is instead scored by calculating the proportion of times a participant responds “more pleasant” and “less pleasant” after each prime.
The AMP has multiple strengths, such as its internal reliability, which is one of the highest among implicit measures (approximately .70-.90; Gawronski and De Houwer, in press), suggesting that results obtained are reliable. The number of trials in an AMP are one of the lowest of all implicit measures, meaning that it is easy and more appropriate to use in short experimental sessions. Priming effects in the AMP are resistant to deliberate attempts to control the influence of the primes – even when the participants are given explicit information about the how the prime stimulus influences their responses (Payne et al., 2005), meaning that this measure is the only measure mentioned that is not susceptible to manipulation. Recently, researchers (e.g. Gawronski and Ye, 2011) have investigated the usefulness of the AMP for semantic associations, broadening its potential applicability.
There are also disadvantages to this measure, such as the relatively flexible response format, which raises questions about what is actually being tested, bringing the measure’s validity into question – Gawronski and Ye (2014) demonstrated that effects in the AMP can be driven by misattribution to both affective and semantic concepts, and Wentura and Degner (2010) ruled out that the effects of the AMP are driven by response priming – both of these findings impact the validity of this measure. This measure cannot be used with participants who are familiar with Chinese ideographs, as the stimuli would no longer be neutral. Participants may also sometimes base responses