People often ask me how fast a response needs to be, to be considered implicit. We always say that implicit processes are fast, but just how fast are they?
But first, let’s ask how fast can people react anyway?
As a teenager, when I got my first digital watch, I used to play with the stopwatch function. You press the same button to start the timer as you do to stop the timer. By pressing the button twice and as fast as possible, you can get a measure of how quickly you can tap your finger. A good score for me was around 0.13 seconds (130 ms). On rare occasions I could get 0.1 (100 ms) and my personal best is 0.09 (90 ms). Remember, the button is a physical button not a virtual button like a touch screen on a mobile device, so takes more to press the button (which could ‘cost’ several milliseconds). So tapping a button as fast as you can (this is not even reacting to anything) can take about 100 ms. In terms of reacting to something, the time it takes to detect whether a target is a word or a nonword (e.g., scrambled letters that don’t make a word) is around 480 ms (Dijksterhuis et al., 1998). So the quickest responses we might expect from any reaction time test might be somewhere between 100 and 500 ms. So turning to implicit responses, we might conclude that fast is 500ms or less and slow (and hence not implicit) is anything else above 500 ms. But this is still just a guess, I haven’t proved anything. In fact, I have spent some time searching academic article databases for an answer to the question ‘How fast is implicit?’ and have yet to read anything this gives me a figure.
But are we even asking the right question?
Some so-called implicit tests make an assumption about this figure, or even avoid the question altogether.
Let’s consider an example from one kind of test in which respondents are asked to make a speeded evaluative response. The question might be ‘Which product do you prefer this one or that one?’ or ‘Which brand do you associate ‘High quality’ with this one or that one?’
The kind of test I have just been describing is a speeded evaluative decision test – henceforth referred to as a Fast Explicit test. The full test goes something like this: it is completed over a number of trials, e.g., 40. Two brands appear on the screen, one to the left and one to the right. A word (some brand attribute, such as ‘Innovative’) appears midway between the two brands and the respondent has to indicate (by way of a key press) which brand the word most applies to in their view. Respondents are encouraged to answer quickly and not think about it too much. So, respondents decide as quickly as they can between the two brands for around 40 word attributes. The computer records the decision and how long it took for the response, measured in milliseconds.
In analysing the results one method is to do a median split on the responses, having removed outliers [a median split involves finding the mid-point of all reaction times and calling anything above this, ‘slow’ and anything below this, ‘fast’] . By doing this, the reaction times are divided into two groups, those below the median reaction time and those above. The median reaction time might be around 750 ms or it might be 2000 ms, depending upon how slow or fast the sample as a whole responded. The data are analysed in two ways – one analysis for responses less than the median value [the ‘fast’ responses] and another one for responses greater than the median value [the ‘slow’ responses]. In fact, given that the median can vary between studies, the terms fast and slow are misnomers, and should instead be ‘faster responses for this sample’ and ‘slower responses for this sample’ – they are not absolute values. The median split method is really a ‘cop-out’ and doesn’t address the ‘how fast is implicit?’ question. To the novel eye, this method appears to provide both implicit and explicit data. But it doesn’t and I will tell you why.
The first reason is that there is no objective measure that tells you the point at which an implicit reaction time becomes explicit. You just have to invent one, just make one up, or, in the above example, let the data decide! Without truly knowing the cut-off point, the figure that indicates the maximum reaction time for an implicit response to be ‘truly’ implicit, you are just making it up. It isn’t scientific at all, it’s purely subjective.
There is an even bigger problem for this fast explicit method, which is that there is no way of knowing whether a respondent on any given trial is trying to respond in the way they have been asked to. There are a number of strategies they could be using to get through the test quickly (such as complete random responses, alternate left brand-right brand responses, or some other combination too difficult to detect by a computer algorithm).
Another problem arises from the fact that most of us are ‘cognitive misers’ – if there is a short cut to a problem then we will take it, rather than the alternative of giving the problem an extensive and tiresome amount of thought. One short cut is to decide that you prefer Brand A over Brand B and so all you need to make sure you select Brand A most of the time. So, overall preference for one brand can lead to a generalised response for that brand.
So where do we go from here? And what is a true implicit test?
True implicit tests are those where there is no evaluative response, there is just a correct answer on each trial – press one key if you see this brand or this kind of word and another key if you see a different brand or a different kind of word. Only one response is correct and you can only proceed if you get it right. The best strategy through this test is to give the correct response on every trial. You might think then that the best strategy would be to press both keys at the same time or both keys as fast as you can. In our tests we detect this kind of key pressing and give warnings or exclude them from our survey. In a true implicit test one compares reaction times under a ‘congruent’ condition with those in an ‘incongruent’ condition, there is no such thing in a fast explicit test.
In a true implicit test, respondents are looking out for ‘targets’, usually two brands. When they see brand A they press one key and when they see brand B they press another key. By flashing word attributes (such as Quality, Trust, and so on) on the screen we try to influence how quickly they are detecting the two brand targets. We can calculate by how much a word might speed up or slow down detection of either target, and hence infer the respondent’s attitudes and feelings to the two brands separately. Now, this is a far more impressive technique than the simple reaction time test of the fast explicit technique. The true implicit test can measure attitudes and feelings without respondent even knowing they are being measured, and without even asking them anything. How do we know this? Because people make spontaneous evaluations very rapidly all of the time. Everything we see, touch, taste, and hear, we very rapidly determine how we feel about it, even without any conscious intention to do so. We can’t help ourselves. Implicit reaction time tests pick this up and in the way I just described.
So, we see that the question ‘How fast is implicit?’ is really the wrong question to ask in the first place, because if you are asking that question then you are actually thinking about a fast explicit test not a true implicit test.
Doctor Implicit, 28th July 2015