Implicit measures of attitudes have been developed since the 1980s, and applied in research on such themes as stereotypes and prejudice (for example, Wittenbrink, Judd & Park, 1997) and health behaviour (Wiers, van Woerden, Smulders & de Jong, 2002; see also Petty, Fazio & Brifiol, 2009; De Houwer & De Bruycker, 2007). Their underlying assumption is that attitudes are associations in the mind (Fazio, Sanbonmatsu, Powell & Kardes, 1986).
Associations that vary in strength form networks, with the stronger ones more easily accessible in our thoughts. Such associated networks are activated automatically. If you find yourself face to face with a shaven-headed man covered in tattoos, wearing sunglasses and accompanied by a dog that has to be muzzled in public, certain thoughts come spontaneously to mind. And the same can happen with members of other groups — people of a particular ethnicity or profession, say — or with a particular product you are familiar with. Implicit measures try to reveal these automatic associations.
The popularity of these measures is due in part to their ability to predict behaviour. For a long time it was assumed that attitudes are the product of explicit reflection, which then guides how we act. In reality, though, much of what we do is more or less habitual and not preceded by a great deal of concerted mental effort (see, for example, Ouellette & Wood, 1998; Wood & Neal, 2007). Strack and Deutsch (2004) state that behaviour can be influenced both by more reflective, reasoned consideration and by impulsive and intuitive processes. Fazio (1990) and Kahneman (2011) also refer to the spontaneous and intuitive bases of attitude versus the more reasoned ones.
So the way we act can be guided by implicit associations as well as more rational processes. Which of these predominates, and to what extent, depends upon the situation and the type of behaviour concerned. Dovidio, Kawakami and Gaertner (2002) found that the degree of friendliness displayed by white test subjects when co-operating with people from other ethnic groups was determined primarily by the associations they had with those groups and not by their explicit attitudes. This and other similar results have prompted great academic interest in implicit measures. Two of the most commonly used techniques are evaluative priming and the implicit association test (IAT).
Evaluative priming was introduced by Fazio, Jackson, Dunton and Williams (1995). It seeks to establish whether, and to what degree, an attitude object affects the categorisation of words as positive or negative. The basic idea is that seeing or hearing one positive word or image makes it easier to identify others in that category and more difficult to identify negative ones. The exact opposite applies when a negative word or image is seen or heard. Fazio, Sanbonmatsu, Powell and Kardes (1986) had already demonstrated that people do indeed respond more quickly when the valence of the ‘prime’ (the attitude object) is the same as that of the ‘target’. Say we show you a picture of a cockroach and then ask you to categorise a word that immediately follows the picture as either positive or negative, as fast as possible. If you are like most people, you find cockroaches repulsive (i.e., it has negative valence). So if the target word following the picture is also negative (`ugly’, ‘dirty’ or ‘disgusting’, say), you are quick to identify it as such. But with incongruent (positive) words like ‘pretty’, ‘tasty’ or ‘attractive’, the process takes just a little longer. By looking at your different response times to the various target words, we can thus measure in a subtle way whether a particular attitude object (in this case a cockroach) elicits positive or negative associations in you. This task can be used to assess the automatic evaluative response to any type of attitude object that can be presented as a prime stimulus such as (single) words or pictures.
The second technique, the IAT, was developed by Greenwald, McGhee and Schwartz (1998). It compares how quickly people are able to categorise specific attitude objects when the categories are coupled with other words or terms, either positive or negative. As an example, imagine we ask you to categorise female names, male names, positive stimuli (party, prize, puppy …) and negative stimuli (murder, bankrupt, rat …). If you have to press one particular key for both female names and positive stimuli and another for both male names and negative stimuli, you will find that a little easier if you have a more positive attitude towards women than men. And it would be slightly more difficult for you if female names are coupled with negative stimuli. In other words, you categorise ‘Mary’ and ‘Lucy’ more quickly in the first of these situations than in the second. This method again exploits the fact that attitude objects (in this case male and female names) trigger automatic evaluative associations that can vary in valence and strength. You can take the test yourself at https:// implicit. harvard. edu.
The IAT is used frequently in research into attitudes and stereotyping (for an overview, see Greenwald, Pohlman, Uhlmann & Banaji, 2009). By replacing the male and female names in the example above with ones associated with a particular ethnic group (Mohammed, Fatima …), for instance, you can gain an insight into the automatic associations evoked in a subject by that group. This is especially useful when explicit, more reasoned measures of stereotyping and prejudice are less predictive of behaviour.
Several implicit measures are now available, most of them reasonably reliable. They are sometimes able to predict behaviour where explicit techniques cannot, particularly when it is habitual, when people have not elaborated thoroughly, or when people are not really aware of their existing preferences (e.g., Galdi, Arcuri & Gawronski, 2008; for a review, see Gawronski, 2009). Implicit measures are now being used in several diverse applied settings, for example in the political domain to predict voting behaviour (Arcuri, Castelli, Galdi, Zogmaister & Amadori, 2008), or to predict consumer decisions (e.g., Dimofte, 2010). This research suggests that both explicit and implicit measures are useful to measure in order to predict behaviour. The specific circumstances within which implicit measures have added value over explicit measures are still in debate though (see for example, Friese, Smith, Plischke, Bluemke & Nosek, 2012; Gawronski, Galdi & Arcuri, 2014).
Notes
Joop van der Pligt & Michael Vliek, The Psychology of Influence: Theory, research and practice (Routledge 2016). ↩