THE STROOP EFFECT: FURTHER TESTS OF THE ATTENTION-CAPTURE HYPOTHESIS
THE STROOP EFFECT: FURTHER TESTS OF THE ATTENTION-CAPTURE HYPOTHESIS
ABSTRACT
Building on the work of Kahneman and Chajcek (1983) and MacLeod and Hodder (1998), this study examined the role of attention in the Stroop effect. A computer-controlled colour-word stimulus presentation and key-press response technique were used to study colour naming response times for dissimilar incongruent word pairs relative to identical incongruent word pairs, and response times for congruent-incongruent word pairs relative to congruent-congruent pairs and incongruent-incongruent pairs. The effect of word position was also investigated. It was found that a stimulus of two different incongruent words produced no more interference than a stimulus of two identical incongruent words and that congruent-incongruent word pairs produced more interference than congruent-congruent pairs but less than incongruent-incongruent pairs. The position of words in the pair had no significant effect. These results contradict the response competition hypothesis and provide further support for the attention-capture hypothesis of Kahneman and Chajcek (1983). However, the factors involved in allocation of attention remain unclear, and it is suggested that these be more thoroughly investigated.
INTRODUCTION
Presented with a stimulus varying on the two dimensions of word and colour, people seem incapable of ignoring the word dimension even when it is irrelevant to the task of naming the colour. This was first demonstrated by Stroop (1935): he gave subjects a list of words printed in different colour inks, and found that while incongruent colours did not interfere with word reading, incongruent words interfered significantly with naming colours, as indicated by longer response times for colour naming. This effect is highly robust: it has been reproduced using a variety of procedures and stimuli, and also seems resistant to practice.
Various factors have been implicated in the Stroop effect: relative discriminability of words and colours (Melara and Mounts 1993) and response modality (McClain 1983), to name but two. However, the two most influential theoretical explanations are the speed of processing and the automaticity hypotheses. Speed of processing models (e.g. Morton and Chambers 1973) assume that words are read faster than colours can be named: when information from each dimension conflicts, the word response, arriving first at the output stage, interferes with the colour response, causing the delay in naming colours. The automaticity hypothesis is based on a distinction between automatic and controlled processes. It proposes that reading words is a strongly automatic process which is involuntary and makes no attentional demands, whereas naming colours is a relatively controlled process, subject to conscious intent and requiring attention. Again, when word and colour responses are different, the more automatic process of word reading interferes with and delays colour naming. It is suggested that word reading is more automatic because we have much more experience at it, an explanation which resembles that put forward by Stroop himself.
Both the speed of processing and the automaticity accounts assign a crucial role to competition between conflicting responses: according to Posner and Snyder (1975, cited in MacLeod 1991), 'the usual Stroop effect arises because of response competition between vocal responses to the printed word and the ink colour'. This response conflict hypothesis also assumes that attention plays no part, with all potential responses being processed.
However, more recent research has tended to contradict this theory. Adopting the Gestaltist distinction between objects and their properties, Kahneman and Henik (1981) propose that during an initial stage of object discrimination attention is allocated to a particular object, and that only the properties of the selected object are processed automatically. This idea is supported by their finding that the Stroop effect is reduced when the incongruent word and ink colour are spatially separate, even though both are equally visible. Kahneman and Chajcek (1983) also found the Stroop effect to be weaker when one colour-neutral word and one conflicting colour word were displayed with a colour bar than when only the conflicting colour word was displayed with the colour bar. They term this the 'dilution effect', and propose an 'attention-capture' hypothesis to explain it. According to this, the processing of words is constrained by capacity limits, and so only the word first attended to will be processed. This runs counter to the assumption of the response competition theory that all possible word responses will be automatically processed without attention being involved. A similar experiment conducted by Yee and Hunt (1991) supports the general idea of attention capture, although their re-analysis of data for individual subjects leads them to suggest that attention is captured at a conceptual level rather than at a perceptual level, as proposed by Kahneman and Chajcek.
The three studies described above all focused on the effect of colour-neutral words and used non-integrated versions of the Stroop stimulus, which tends to produce less interference. MacLeod and Hodder (1998) examined whether their results could be generalised to a standard integrated Stroop stimulus. They found that presenting two different incongruent colour words did not cause significantly more interference in colour naming than two identical incongruent words, and so reject the response conflict view which predicts that more potential responses would create more competition and hence more interference. Like Kahneman and Henik (1981) and Kahneman and Chajcek (1983), ...
This is a preview of the whole essay
The three studies described above all focused on the effect of colour-neutral words and used non-integrated versions of the Stroop stimulus, which tends to produce less interference. MacLeod and Hodder (1998) examined whether their results could be generalised to a standard integrated Stroop stimulus. They found that presenting two different incongruent colour words did not cause significantly more interference in colour naming than two identical incongruent words, and so reject the response conflict view which predicts that more potential responses would create more competition and hence more interference. Like Kahneman and Henik (1981) and Kahneman and Chajcek (1983), they conclude that attention is involved at an early stage in processing and that 'the first word processed captures attention, thereby screening any others out'.
The present study constitutes a further investigation of this attention-capture hypothesis. It aims firstly to reproduce the basic Stroop effect, and then to replicate the finding of MacLeod and Hodder (1998) that no more interference will be found for two different incongruent words than for two identical incongruent words. It also proposes to examine whether the 'dilution effect' observed by Kahneman and Chajcek (1983) with non-integrated stimuli and neutral non-colour words can be generalised to integrated stimuli using two colour words, one congruent and one incongruent. If so, it would be expected that a stimulus comprising one incongruent and one congruent word would produce more interference than a stimulus of two congruent words and less interference than a stimulus of two incongruent words. Finally, the possible effects of word position in capturing attention will be considered.
METHOD
Design
A related measures design was used.
There were five experimental conditions, presented in random order:
* both congruent - e.g. blue above blue, printed in blue
* incongruent same - both words incongruent and identical; e.g. blue above blue, printed in red
* incongruent different - both words incongruent and different; e.g. blue above red, printed in green
* top incongruent - top word incongruent, bottom word congruent; e.g. blue above red, printed in red
* bottom incongruent - top word congruent, bottom word incongruent; e.g. blue above red, printed in blue.
Thus the independent variables manipulated were type of word (incongruent / congruent), similarity of words (identical / different), number of incongruent words (one / two) and position of incongruent word (top / bottom).
The dependent variable in all conditions was the time taken to name the ink colour, measured in milliseconds.
Subjects
Subjects were 76 first year psychology students at Oxford Brookes University. No controls for visual acuity or colour blindness were applied. Data from one subject were excluded on the basis of an error rate of 25%. Data from 75 subjects remained for analysis.
Materials
Stimulus presentation and response were controlled by a computer system.
The background of the computer screen was black; all instructions and fixation stimuli were presented in white.
Stimulus materials consisted of four colour words - red, blue, yellow and green - presented in pairs, one above the other in the centre of the screen. Words were printed in these same four colours.
Responses were made using a key-press technique. Red, blue, green and yellow stickers were placed over the numbers 8, 4, 6 and 2 on the computer's numeric key-pad.
Procedure
Subjects were tested over four days in groups of 18 - 20.
The following display sequence was adopted for each trial:
* a 'next trial' warning for 1 second
* a blank screen for 250 ms.
* a fixation stimulus (XXXXXX) for 250 ms.
* a blank screen for 250 ms.
* the word pair stimulus (displayed until response).
Each subject experienced eight practice trials and sixty experimental trials: there were twelve trials for each of the five experimental conditions, presented in random order. In all trials subjects were asked to identify the colour the words were printed in. Subjects were instructed to use only their index finger to press the response keys, and to keep this finger on the number 5 key (the mid-point between all response keys) between trials. This ensured that all keys were visible and could be reached equally quickly.
The median response time for each subject in each condition was calculated: this measure was preferred to the mean as it is less affected by outlying results.
Results were analysed using the related t-test.
RESULTS
Table 1: means and standard deviations for the five experimental conditions
CONDITION
MEAN RESPONSE TIME (MILLISECONDS)
STANDARD DEVIATION
) Both congruent
696? 3
93? 994
2) Incongruent same
743? 76
01? 225
3) Incongruent different
739? 107
99? 491
4) Top incongruent
709? 113
97? 244
5) Bottom incongruent
710? 247
23? 213
It can be seen from the above that response times were lowest when both words were congruent. Response times were highest when both words were incongruent, and mean response times for conditions 2 and 3 were fairly similar. Approximately intermediate response times were produced when one word only was incongruent, and mean response times for conditions 4 and 5 were very similar.
Table 2: summary of t-test results
HYPOTHESIS
CALCULATED T
TABULATED T
RH1
6? 28
? 666
RH2
-0? 592
? 666
RH3
2? 076
? 666
RH4
-6? 028
-1? 666
RH5
0? 145
? 993
Research hypothesis 1
Tabulated value of t (p < 0? 05, df = 74, one-tailed test) = 1? 666
Calculated value of t = 6? 28 (significant at the p < 0? 005 level)
As 6? 28 > 1? 666 and the difference is in the predicted direction, null hypothesis 1 is rejected and research hypothesis 1 supported.
Response times for two identical incongruent words were significantly greater than response times for two congruent words.
Research hypothesis 2
Tabulated value of t (p < 0? 05, df = 74, one-tailed test) = 1? 666
Calculated value of t = - 0? 592
As - 0? 592 < 1? 666, research hypothesis 2 is rejected and null hypothesis 2 retained.
Response times for two different incongruent words were not significantly greater than response times for two identical incongruent words.
The difference between response times was in the opposite direction to that predicted by RH2, although this tendency was slight and not statistically significant.
Research hypothesis 3
Response times for congruent-incongruent word pairs were obtained by combining data from the 'top incongruent' and 'bottom incongruent' conditions.
Tabulated value of t (p < 0? 05, df = 74, one-tailed test) = 1? 666
Calculated value of t = 2? 076
As 2? 076 > 1? 666 and the difference is in the predicted direction, null hypothesis 3 is rejected and research hypothesis 3 supported.
Response times for a congruent-incongruent word pair were significantly greater than response times for a congruent-congruent word pair.
Research hypothesis 4
Response times for congruent-incongruent word pairs were obtained by combining data from the 'top incongruent' and 'bottom incongruent' conditions.
Response times for incongruent-incongruent word pairs were obtained by combining data from the 'incongruent same' and 'incongruent different' conditions.
Tabulated value of t (p < 0? 05, df = 74, one-tailed test) = - 1? 666
Calculated value of t = - 6? 028 (significant at the p < 0? 005 level)
As - 6? 028 > - 1? 666 and the difference is in the predicted direction, null hypothesis 4 is rejected and research hypothesis 4 supported.
Response times for congruent-incongruent word pairs were significantly less than response times for incongruent-incongruent word pairs.
Research hypothesis 5
Tabulated value of t (p < 0? 05, df = 74, two-tailed test) = 1? 993
Calculated value of t = 0? 145
As 0? 145 < 1? 993, research hypothesis 5 is rejected and null hypothesis 5 retained.
There was no significant difference between response times when the top word only was incongruent and response times when the bottom word only was incongruent.
DISCUSSION
The classic Stroop effect predicted by hypothesis 1 was observed, with colour naming significantly affected by incongruent colour words. This successful reproduction of this well-documented effect serves as a valuable indication that other results obtained may be considered valid and comparable with the findings of previous studies. This point is perhaps especially important as the present study differs from most prior research in using manual rather than vocal responses, and response mode is considered by some to have an important impact on the Stroop effect. For example, McClain (1983) found that the Stroop effect was eliminated with the type of manual response mode employed here.
Response times for two different incongruent words were not significantly longer than for two identical incongruent words. This finding is in accordance with the results obtained by MacLeod and Hodder (1998), and supports their rejection of the response competition hypothesis, which predicts that a greater number of possible word responses will lead to more competition and so more interference. It was also found that word pairs consisting of one congruent and one incongruent word cause more interference than a pair of congruent words but less interference than a pair of incongruent words. This indicates that the so-called dilution effect observed by Kahneman and Chajcek (1983) for non-integrated stimuli and colour-neutral words can be generalised to integrated stimuli comprising two colour words.
Taken together, these findings provide further support for the attention-capture hypothesis proposed by Kahneman and Chajcek (1983) and upheld by MacLeod and Hodder (1998). It seems that only one word of the pair is attended to and processed, and thus that, contrary to the assumptions of the automaticity and response competition accounts, reading on the Stroop task is not entirely involuntary and free of attentional demands: an initial stage of object discrimination appears to be involved.
What factors might be involved in this object discrimination remain unclear, however. Whether the top word or bottom word was incongruent had no significant effect on response times, suggesting that word position is not a crucial factor in the allocation of attention. (It might, for example, have been supposed that the fact that we read from top to bottom could have some effect). This question is deserving of a more sustained examination. It has been briefly considered by Yee and Hunt (1991), who concluded that word length and the spatial allocation of attention were not important factors. A starting point for further investigation of such factors might be to repeat the present experiment but to analyse data on a within-individual level rather than averaging it across individuals. It is possible that, as Yee and Hunt claim, a study of individual differences may provide valuable information about the attentional mechanisms involved in the Stroop effect.
In conclusion, the results of the present study provide further evidence against a response competition account of the Stroop effect, which argues that all words are involuntarily and automatically processed, and support for the theory that only one value on each dimension captures attention. However, the study offers no more information as to what influences attention capture, a question which merits further examination.
APPENDIX 1
REFERENCES
KAHNEMAN, D. & CHAJCEK, D. (1983). Tests of the automaticity of reading: dilution of Stroop effects by color-irrelevant stimuli. Journal of Experimental Psychology: Human Perception and Performance 9, 497 - 509.
KAHNEMAN, D. & HENIK, A. (1981). Perceptual organisation and attention. In M. Kubovy & J.R. Pomerantz (Eds.), Perceptual organisation. Hillsdale, N.J.: Erlbaum.
McCLAIN, L. (1983). Effect of response type and set size on Stroop color-word performance. Perceptual and Motor Skills 56, 735 - 743.
MacLEOD, C.M. (1991) Half a century of research on the Stroop effect: an integrative review. Psychological Bulletin 109, 163 - 203.
MacLEOD, C.M. & HODDER, S.L. (1998). Presenting two incongruent color words on a single trial does not alter Stroop interference. Memory & Cognition 26 (2), 212 - 219.
MELARA, R.D. & MOUNTS, J.R.W. (1993). Selective attention to Stroop dimensions: effects of baseline discriminability, response mode, and practice. Memory & Cognition 21, 627 - 645.
MORTON, J. & CHAMBERS, S.W. (1973). Selective attention to words and colours. Quarterly Journal of Experimental Psychology 25, 387 - 397.
STROOP, J.R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology 18, 643 - 622.
YEE, P.L. & HUNT, E. (1991). Individual differences in Stroop dilution: tests of the attention-capture hypothesis. Journal of Experimental Psychology: Human Perception and Performance 17 (3), 715 - 725.
APPENDIX 2
SAMPLE CALCULATION OF RELATED T-TEST, USING CASES 1 - 20 FOR THE 'TOP INCONGRUENT' AND 'BOTTOM INCONGRUENT' CONDITIONS
Subjects
Response times:
top incong. (x)
Response times:
bottom incong. (y)
Difference
d = (x - y)
d2
794
875
- 81
6561
2
792
756? 5
35? 5
260? 25
3
685? 5
626
59? 5
3540? 25
4
630
656? 5
- 26? 5
702? 25
5
837? 5
768? 5
69
4761
6
609
714
- 105
1025
7
656
705? 5
- 49? 5
2450? 25
8
606
589? 5
6? 5
272? 25
9
927? 5
922? 5
5
25
0
781
727? 5
53? 5
2862? 25
1
580? 5
531? 5
49
2401
2
630
587? 5
42? 5
806? 25
3
671
671
0
0
4
981
967? 5
3? 5
82? 25
5
735? 5
780
- 44? 5
980? 25
6
744? 5
632? 5
12
2544
7
712
714
- 2
4
8
724? 5
694
30? 5
930? 25
9
644
624
20
400
20
668? 5
668? 5
0
0
Totals:
-
-
? d = 198
? d2= 53707? 5
t = 0? 848
df = n - 1 = 19
Tabulated value of t (p < 0? 05, df = 19, two-tailed test) = 2? 093
0? 848 < 2? 093: thus in this example no significant difference would be found.
Appendix 2 contd.
STATISTICAL CALCULATIONS
AL-Stats Stroop 7703 class June 2000 page 1
(AL-Stats Version 6.15) Monday June 19 2000, at 11:01 am
Registered to Oxford Brookes Psychology
Calculation of Paired t-test on Stroop 7703 class June 2000: hypothesis 1
Columns used: C3 = Incongruent Same, C6 = Both Congruent
Test the alternative hypothesis :Incongruent Same > Both Congruent
Required significance level = 5.0%
The critical value of t (0.050,74) = 1.666
The calculated value of the test statistic t = 6.280
Since 6.280 >1.666 Reject the Null Hypothesis
Calculation of Paired t-test on Stroop 7703 class June 2000: hypothesis 2
Columns used: C2 = Incongruent Different, C3 = Incongruent Same,
Test the alternative hypothesis :Incongruent Different > Incongruent Same
Required significance level = 5.0%
The critical value of t (0.050,74) = 1.666
The calculated value of the test statistic t = -0.592
The observed difference is in the other direction from
the predicted hypothesis
Do NOT Reject the Null Hypothesis
Calculation of Paired t-test on Stroop 7703 class June 2000: hypothesis 3
Columns used: C6 = Both Congruent, C8 = One Incongruent
Test the alternative hypothesis :One Incongruent > Both Congruent
Required significance level = 5.0%
The critical value of t (0.050,74) = 1.666
The calculated value of the test statistic t = 2.076
Since 2.076 >1.666 Reject the Null Hypothesis
Calculation of Paired t-test on Stroop 7703 class June 2000: hypothesis 4
Columns used: C7 = Both Incongruent, C8 = One Incongruent
Test the alternative hypothesis :One Incongruent < Both Incongruent
Required significance level = 5.0%
The critical value of t (0.050,74) = -1.666
The calculated value of the test statistic t = -6.028
Since -6.028 <-1.666 Reject the Null Hypothesis
Calculation of Paired t-test on Stroop 7703 class June 2000: hypothesis 5
Columns used: C4 = Top Incongruent, C5 = Bottom Incongruent
Test the alternative hypothesis :Top Incongruent <> Bottom Incongruent
Required significance level = 5.0%
The critical value of t (0.050,74) = 1.993
The calculated value of the test statistic t = 0.145
Since Magnitude of 0.145 <1.993 Do NOT Reject the Null Hypothesis