kuhn 2008 retrospectivefreechoice

IntentionReconstructionBack referral

ing a stop signal paradigm and an intentional action paradigm we show that participants

eclared wconc

and this predictable component is removed from the incoming sensory signal, thereby causing the attenuation (Bays, Flan-agan, & Wolpert, 2006).

1053-8100/$ - see front matter 2008 Elsevier Inc. All rights reserved.

* Corresponding author. Address: Ghent University, Faculty of Psychology and Educational Sciences, Department of Experimental Psychology, HenriDunantlaan 2, 9000 Gent, Belgium. Fax: +32 9 264 6403.

E-mail address: [email protected] (S. Khn).

Consciousness and Cognition xxx (2008) xxxxxx

Contents lists available at ScienceDirect

Consciousness and Cognition

journal homepage: www.elsevier .com/locate /concog

ARTICLE IN PRESSdoi:10.1016/j.concog.2008.09.007heard about the experiments of Benjamin Libet claiming that actions are unconsciously initiated and are already predictablefrom external electrophysiological measures before we ourselves report consciousness of the intention to act (Libet, Gleason,Wright, & Pearl, 1983). This implies that the feeling of being in control of an action has to be a reconstructed experience (of-ten called subjective back-referral).

An alternative hypothesis to this retrospective reconstruction assumes a predictive mechanism of phenomenal experi-ence of intention. Wolpert assumes that a forward model makes predictions about the behavior of the motor system andits sensory consequences (Wolpert, Ghahramani, & Jordan, 1995). Those predictions are used to compare the actual outcomeof a motor command with the desired outcome enabling rapid error correction before sensory feedback is available. In linewith that model sensory attenuation has been shown to result from these kinds of predictive mechanisms. When a motorcommand is generated, an efference copy of this command is used to predict the sensory consequences of the actionFree willConsciousness

1. Introduction

Although Sigmund Freud (1917) das if we were. We might have arrangnot seem to have inuenced our self-Please cite this article in press as: Khn, Sness and Cognition (2008), doi:10.1016/j.sometimes indicate to have intentionally initiated an action while reaction time datastrongly suggest that they in fact failed to stop the action. In a second experiment we dem-onstrate that the number of trials in which participants misattributed their awareness ofintention varied with the intentional involvement during action planning. Our data supportthe retrospective account of intentional action. Furthermore, we introduce an experimentalapproach that objecties introspective judgments of awareness of intention.

2008 Elsevier Inc. All rights reserved.

ed we are not der Herr im eigenen Haus (master of our own house), we still feelith the knowledge of unconscious motives we do not have access to, but this doeseption of having primary access to our intentions and action plans. We might haveRetrospective construction of the judgement of free choice

Simone Khn a,b,*, Marcel Brass b

aDepartment of Psychology, Max Planck Institute for Human Cognitive and Brain Sciences, Stephanstr. 1A, 04103 Leipzig, GermanybGhent University, Faculty of Psychology and Educational Sciences, Department of Experimental Psychology, Henri Dunantlaan 2, 9000 Gent, Belgium

a r t i c l e i n f o

Article history:Received 4 April 2008Available online xxxx

Keywords:Voluntary action

a b s t r a c t

The problem of free will lies at the heart of modern scientic studies of consciousness.Some authors propose that actions are unconsciously initiated and awareness of intentionis referred retrospectively to the action after it has been performed [e.g. Aarts, H., Custers,R., & Wegner, D. M. (2005). On the inference of personal authorship: Enhancing experi-enced agency by priming effect information. Consciousness & Cognition, 14, 439458]. Thiscontrasts with the common impression that our intentions cause those actions. By combin-.& Brass, M. Retrospective construction of the judgement of free choice. Conscious-concog.2008.09.007

2 S. Khn, M. Brass / Consciousness and Cognition xxx (2008) xxxxxx

ARTICLE IN PRESSNevertheless these predictive mechanisms are not capable to explain the results of illusory control studies. Wegner e.g.uncovers a reconstructive mechanism of experience of intention by demonstrating that subjects can be led to feel that theyconsciously intended actions or consequences of actions they did not produce themselves (Aarts, Custers, & Wegner, 2005).This phenomenon is said to be based on the mechanism of back referral of an intention. In a study by Wegner and Wheatley(1999) subjects retrospectively attribute conscious intentions to themselves in order to explain actions that were actuallyperformed by another person, but could have been performed by them. In a study about magical thinking participants wereled to belief that they harmed another person via voodoo hey (Pronin, Wegner, McCarthy, & Rodriguez, 2006). Recently it hasbeen shown that also plagiarism can be induced by means of misattribution of effort cues (Preston &Wegner, 2007). In mostof these studies the illusion of will is evoked in a context where externally produced action effects are attributed to the self.The question is, does this confusion of intention only occur between the effects of self-generated and externally-generatedactions, or are there also confusions about the voluntariness of our own actions? Wegner and his colleagues have shown thatparticipants can be led to ascribe an intention to an action that they did not initiate themselves. But what has not been dem-onstrated is that participants ascribe intentions to their actions, although they did not actually intend them.

In our study, we address this by asking the following question: Are subjects able to tell the difference between a voluntarydecision to resume an ongoing action and an inability to stop an ongoing action?

In order to investigate this question, we used a modication of the classical stop task (Logan & Cowan, 1984). In stop tasksparticipants are usually engaged in a reaction time task (the primary response task), and occasionally and unpredictably,they are confronted with a signal (often a tone or a colour change), that instructs them to inhibit their response to the rststimulus. Since the subjects usually have to respond (stop trials only occur in 2025% of the trials), responding is so to speakthe default mode in the task. In contrast to a Go/NoGo paradigm, the stop signal usually occurs at one of several delaytimes following the presentation of the primary response stimulus. In order to create a condition of free choice, we modiedthe classical stop task by introducing a second signal in addition to the classical stop signal. We call this signal the decidesignal, which assigns the task to choose between executing the prepared action and refraining from the prepared action.After those trials we asked subjects whether they were able to decide or whether they responded impulsively to the primaryresponse stimulus.

We aimed at comparing the reaction time (RT) in decision trials in which the subjects chose voluntarily to press the but-ton with RTs in primary response trials in order to explore whether subjects are able to discriminate between acting in de-fault mode without being able to stop and deciding voluntarily to resume the prepared action. If participants are capable ofdistinguishing those states we should nd no decide trials in which subjects state to have chosen voluntarily to resume theprepared action in the range of primary response RTs, because the process of stopping an ongoing action and reinitiating itvoluntarily should take time. Whensoever we nd decide trials which participants acknowledge as successful decision trialsin the range of primary RTs, we have to assume that they are not able to reliably discriminate between producing an actionwithout stopping and stopping an action before voluntarily resuming.

2. Experiment 1

2.1. Methods

2.1.1. ParticipantsTwelve subjects (4 male, 8 female, mean age: 24.7 years) were recruited and were paid 14 Euro for participation in two

experimental sessions. They had normal or corrected-to-normal vision. None of the subjects had a history of neurological,major medical, or psychiatric disorder. All were right handed, as assessed by the Edinburgh Inventory (Oldeld, 1971).

2.1.2. Apparatus and stimuliStimuli of the task were the uppercase letters M, V, N, and W, with a visual angle of 0.4 at a distance of 80 cm from the

screen, presented in black on a light grey background. The signal to stop or decide was a change of letter colour (black to pinkor black to blue) counterbalanced across subjects.

2.1.3. ProcedureWe adopted the choice RT task used by De Jong, Coles, Gratton, and Logan (1990), De Jong, Coles, and Logan (1995) in

which the stimulus letters M and V required a left button press with the right index nger; the letters N and W to a rightbutton press with the right middle nger. Each trial started with the presentation of a xation cross on the centre of thescreen for 500 ms. Then it disappeared and was followed by one of the black stimulus letters after 500 ms. In 25% of the trialsthe black stimulus was coloured after a signal delay (SD). The black or coloured letter was displayed on the screen until thetrial ended after 2000 ms or until a response was given (Fig. 1). After that the screen was blank for 1000 ms.

For half of the subjects a change to blue meant stop and a change to pink meant decide, whereas for the other half themapping was reversed.

After the so called decide trials, in which participants were prompted to choose between performing and not-performingthe prepared action, another screen was presented asking whether the subject did have the chance to decide or reacted in thedefault mode in response to the black stimulus (Decided? yes - no in German: Entschieden? ja - nein). The writtenPlease cite this article in press as: Khn, S.& Brass, M. Retrospective construction of the judgement of free choice. Conscious-ness and Cognition (2008), doi:10.1016/j.concog.2008.09.007

S. Khn, M. Brass / Consciousness and Cognition xxx (2008) xxxxxx 3

ARTICLE IN PRESSinstruction for the decide condition read If you succeed not to respond impulsively when the letter changes its colour toblue/pink decide between pressing or not pressing the button at this point in time. Decide as if you were ipping a coin with-out having a certain rule in mind (e.g. alternating between pressing and not pressing). Participants were instructed to decideabout equally often to execute or not to execute the response in decide trials. In order to avoid waiting strategies, the pri-

Fig. 1. Schematic drawing of the different conditions in Experiment 1.macy of the choice RT task was emphasized in the instruction and the participants were informed that the probability ofsuccessful stopping or deciding would approximate to .50, irrespective of whether they were postponing their responseor not (The main task is to respond as fast as possible to the letters presented on the screen without making a lot of mis-takes. Do not wait with your response until the letter changes colour.)

Our design therefore includes six different conditions: the primary response, stop, failed-to-stop, decide-go, decide-nogoand failed-to-decide condition.

For the rst stop and decide trial we used a SD of 300 ms. Afterwards the SD was continuously modied according to astaircase procedure. If the subject succeeded in withholding the response, the SD increased by 20 ms, making the task moredifcult; conversely, if they failed, the SD decreased by 20 ms, making the task easier. In case of decide trials the staircasewas treated separately. If the subjects were successful in decide trials and indicated with a button press that they chose theirresponse or no-response intentionally, the staircase was prolonged by 20 ms; if they were not successful and indicated thatthey did not decide intentionally, it was shortened by 20 ms. This staircase procedure was applied in order to achieve a prob-ability of approximately .50 successful and nonsuccessful trials. In order to allow an optimal variation of the signal delay weused the rather difcult stimulus-response mapping (MV = index nger, NM = middle nger).

A training block at the beginning of the experiment contained 30 trials and was discarded from further analysis. Then veblocks with 96 trials each followed (72 primary response trials, 12 decide trials, 12 stop trials per block). The rst sessionended with a questionnaire concerning strategies and questions concerning the task (Did you have the feeling that you de-layed your response in order to make sure the letter is not changing its colour? Did you have any strategies?).

The second session also started with a training block, and was again followed by ve experimental blocks. It ended withthe completion of a questionnaire containing the MarloweCrowne Social Desirability Scale (1960) (in the German transla-tion by Dickenberger, Holtz, & Gniech, 1978), the Paulhus Balanced Inventory of Desirable Responding (1994) (in the trans-lation by Musch, Brockhaus, & Brder, 2002) and the I7 by Eysenck, Daum, Schugens, and Diehl (1990). Each session lastedapproximately 40 min.

2.1.4. Data analysisVisual inspection of the RT-distributions of the participants revealed a bimodal distribution of decide-go trials with one

peak underneath the primary response distribution (Fig. 2). In order to substantiate this pattern statistically we used anexpectation-maximization (EM)-algorithm to t two lognormal distributions to the data of all subjects.

Please cite this article in press as: Khn, S.& Brass, M. Retrospective construction of the judgement of free choice. Conscious-ness and Cognition (2008), doi:10.1016/j.concog.2008.09.007

We split the decide-go trials into two categories early decide-go and late decide-go trials (see Fig. 3 for an overview ofthe different trial types). We dened early decide-go trials as those trials underneath the left curve up to the point where 10%of the area under the right curve is cut off (left vertical line in Fig. 4). Similarly late decide-go trials were dened to be those

Fig. 2. RT distributions of primary response, failed-to-decide and decide-go trials over all participants (the decide-go, failed-to-decide trials are arbitrarilyrescaled by multiplication with the number of primary response trials and division by factor 2).


ARTICLE IN PRESSunderneath the right curve up to the point where 10% of the area under the left curve is cut off (right vertical line in Fig. 4).The trials in between the early and late trials where not analyzed, because a clear assignment was not possible.

In order to obtain a measure for the goodness of t we compared the empirical cumulative distribution function (cdf) ofthe raw data with the cdf of the tted data using KolmogorovSmirnov-Test (KS-test). Similar KS-tests were performed totest differences in distributions of early decide-go trials and the tted left-sided lognormal distribution, same as differencesbetween late decide-go trials and the tted right-sided lognormal distribution separately. For the KS-tests we randomly as-signed the decide-go trials to the early or late category according to probabilities based on the proportion of the two ttedcurves. Since our classication of the raw data was therefore not independent of the tting we repeated the assignment pro-cedure as well as the related KS-tests 1000 times and averaged the resulting Dk values to allow for random variations.

KS-tests comparing the tted distributions with the raw data did not reject the null hypothesis (both distributions are thesame), neither for the bimodal tted distribution when compared to the raw data (Dk = 0.040; p = .56), nor for the tted earlyFig. 3. Scheme of the different classes of decide trials.



ARTICLE IN PRESSdecide-go trials when compared to the raw data (Dk = 0.041; p = .97) and the late decide-go trials compared to the raw data(Dk = 0.058; p = .39) indicating a good t of the bimodal curve. A comparison of the raw data distribution of early decide-gotrials compared to the distribution of late decide-go trials we found a highly signicant difference in distribution (Dk = 0.947;p < .001). Those tests reveal that our data is appropriately represented by a bimodal reaction time distribution and that thereis a pronounced difference between early and late decide-go trials.

We used user-written functions based on MATLAB 7 (MathWorks, Inc., 2004), for processing the data.

2.1.5. Results and discussionThe mean percentage of successful inhibition trials in the stop condition (52%) equalled approximately the expected prob-

ability of .50, which indicates a proper delay assignment according to Logan (1995). The mean percentage of trials in whichthe subjects indicated an intentional decision in the decide condition was also in that range (54%). The distribution of decide-go trials and decide-nogo trials are also nearly equal (52.5 vs. 47.5%). The number of wrong choice responses in the primaryresponse task was low with a mean of 4.2%.

The means across median RTs are presented in Table 1.In the questionnaire after the rst session no participant reported the feeling of delaying their response in order to make

sure that the letter does not change its colour indicating compliance with the task instruction. No participant mentioned thestrategy to build a prior intention whether to act or not if a decide signal occurs.

Since we set out to explore whether subjects are able to discriminate between acting without being able to stop a pre-pared action and deciding voluntarily to resume the prepared action we compared the reaction times in decide-go trialsin which the subjects chose voluntarily to press the button with primary response trials.

The data reveal a bimodal distribution of the decide-go trials, of which the rst peak is situated approximately under-neath the median of the primary RTs, whereas the second peak is about 600 ms later (see Fig. 2).

With an EM-algorithm we tted two lognormal distributions to the RT distribution of decide-go trials (Fig. 5). In order toconrm this twofold pattern we performed KS-tests revealing that the data is appropriately represented by a bimodal reac-tion time distribution.

Fig. 4. Schematic drawing of the 10% criterion used to assign trials to the early decide-go vs. late decide-go category.Using the 10% criterion, on average ten trials belong to the category of early decide-go trials and 19 trials to late de-cide-go trials per participant (t(11) = 1.31, p > .05). As expected the mean RTs of early decide-go trials (757 ms) differedsignicantly from the mean RTs of late decide-go trials (1306 ms). There was no signicant difference between failed-to-de-cide trials in which participants report that they were not able to inhibit the ongoing response in order to choose betweenresponding or not and early decide-go trials (t(11) = .185, p > 0.05).

Now, why do those early decide-go trialsthe false alarmsoccur? One could assume that participants are lying. Subjectscould, for strategic reasons and/ or because they want to perform well, indicate that they decided although they did not vetoin order to decide whether to act or not to act afterwards. The lack of signicant correlations between the amount of earlydecide-go trials and the MarloweCrown Social Desirability Scale (M = 16.08, SD = 1.72; r = .442, p > .05; the lying or social

Table 1Mean of median reaction times in milliseconds (ms)

Experiment 1 Experiment 2 stimulus condition Experiment 2 intention condition

Primary response trials 694 556 587Failed-to-stop trials 617 488 517Failed-to-decide trials 605 506 542Decide-go trials 1101 861 995



ARTICLE IN PRESSdesirability hypothesis would predict a positive correlation) or the Paulhus Balanced Inventory of Desirable Responding(impression-management: M = 39.17, SD = 4.78; r = .284, p > .05; self-deception: M = 33.08, SD = 7.22; r = .046, p > .05)seems to argue against that.

A more profound argument against the lying hypothesis is that the process of lying after early decide-go trials should taketime, and should make the response to the Decided?request more difcult. But a comparison of the time the subjectsneeded to respond to the Decided?request at the end of decide trials revealed a signicant difference between the decidetime following early decide-go trials compared to late decide-go trials pointing into the opposite direction (t(11) = 3.095,p < .05). The decision after early decide-go trials (579 ms) was 81 ms shorter compared to the decision after late decide-gotrials (660 ms). If one assumes that a short reaction time is a marker for condence, the subjects are even more condent inthe early decide-go trials compared to the late decide-go trials.

Another alternative explanation could be that early decide-go trials are trials in which the participants accidentallypressed the wrong button in response to the request screen (Decided?) where they indicated whether they decided ornot. This would predict a positive correlation of the number of early decide-go trials with the impulsiveness measure. Butwhat we nd is a signicant negative correlation of the number of early decide-go trials and I7 (M = 1.52, SD = .21;r = .744, p < .01). This result suggests that subjects with higher impulsivity produced less early decide-go trials.

An incidental observation was, that when we showed participants their own data of the rst session at the end of thesecond session and explained that the early decide-go trials are presumably trials in which they did not have the time todecide between responding and not-responding, subjects reacted with astonishment. They afrmed that they only indicatedyes in response to the Decided?request when they actually had the clear-cut impression that they rst stopped anddecided afterwards.

Another alternative explanation for the occurrence of early decide-go trials could be that participants form a prior inten-tion about how they will respond in the next trial they are given the chance. But this is highly implausible because this strat-egy does not explain the occurrence of late decide-go trials and the presence of a bimodal distribution of the RTs.

Taken together these results suggest that our participants were convinced that they stopped the prepared action and

Fig. 5. Lognormal tted reaction time distribution of decide-go trials in Experiment 1 (The y-axis represents a normalized scale setting the area under thetted early & late decide-go curve to 1).went through a decision process in the early decide-go trials but actually did not. In order to nd factors that inuencethe amount of early decide-go trials we varied the deliberateness of the primary response. Our prediction was that for freelychosen primary responses it should be easier to distinguish whether the action has been stopped or not compared to stim-ulus-driven primary responses. In other words: Subjects should be more condent in knowing whether they stopped an ac-tion, when they chose what act to perform in the rst place.

3. Experiment 2

3.1. Methods

3.1.1. ParticipantsTwelve subjects (6 male, 6 female, mean age: 24.3) were recruited for this behavioural study, none of them having par-

ticipated in the aforementioned experiments. They were paid 28 Euro for participation in four experimental sessions.

3.1.2. Apparatus and stimuliStimuli were black dots presented in front of a light grey background, either located in the middle, on the right or on the

left side of the screen, with a visual angle of 0.35. The participants were seated at a distance of 80 cm in front of a screen. Thesignals to stop or decide were changes of dot colour (black to pink or black to blue).


See Experiment 1 for further information.


ARTICLE IN PRESSThe KS-tests comparing the tted distributions with the raw data did not reject the null hypothesis (both distributions arethe same), neither for the bimodal tted distribution when compared to the raw data (stimulus condition: Dk = 0.030;p = .72; intention condition: Dk = 0.051; p = .16), nor for the tted early decide-go trials when compared to the raw data(stimulus condition: Dk = 0.074; p = .35; intention condition: Dk = 0.085; p = .29) and the late decide-go trials compared tothe raw data (stimulus condition: Dk = 0.045; p = .42; intention condition: Dk = 0.064; p = .10). In a comparison of the rawdata distribution of early decide-go trials compared to the distribution of late decide-go trials we found a signicant differ-ence in distribution (stimulus condition: Dk = 0.923; p < .01, intention condition: Dk = .934; p < .01). The KS-tests support thatthe data is appropriately represented by a bimodal reaction time distribution.

3.1.5. Results and discussionThe means across median RTs are presented in Table 1.With this experiment we were able to replicate the bimodal pattern of the decide-go RT distribution. The tting results

are displayed in Fig. 6a and b.This is a replication of the results in Experiment 1 implying that this data is again appropriately represented by a bimodal

reaction time distribution. Likewise both conditions in Experiment 2 reveal no signicant difference between failed-to-de-cide trials and early decide-go trials (stimulus condition: t(11) = 2.196, p > .05; intention condition: t(11) = 1.802, p > .05).

Since the experiment was mainly targeted at investigating the inuence of different degrees of deliberateness of the pri-mary response onto the number of early decide-go trials, we calculated the ratio of the number of early decide-go trials andthe sum of early and late decide-go trials for the stimulus and intention condition separately (#early/(#early + #late)). As pre-dicted a paired t-test revealed a signicant difference between both ratios (t(11) = 2.236, p < .05), with a larger ratio for thestimulus condition compared to the intention condition. This result conrms that subjects aremore condent in their decisionratings when they voluntarily initiate the action compared to when they react in response to an external stimulus. Addition-ally this modulation of the number of early decide-go trials within-subject argues against the lying hypothesis, because there isno reason to believe that the tendency to respond dishonestly should depend on the deliberateness of the primary response.

Recapitulatory, the results of Experiment 2 replicate the bimodal pattern of the decide-go RT distribution and demon-strate that reactions in response to an indicative stimulus (stimulus condition) are more prone to confusion concerningthe inhibition and the voluntary continuation of the action in comparison to reactions performed after a free choice stimulus(intention condition). Therefore our hypothesis that participants perform better in determining whether they stopped an ac-tion, when they initially chose to perform that action compared to when they had to respond to an external stimulus, hasbeen conrmed.

4. General discussion

The aim of the present study was to investigate retrospective construction of intentions in a context of a stop signal par-adigm with an additional condition that required participants to decide whether to execute the pre-specied response ornot. Experiment 1 indicated that participants reported a substantial amount of trials as being intentionally decided that wereactually in the reaction time range of the primary response time. This nding strongly suggests that they falsely attributed anintentional decision to act when they were actually not able to stop. We conrm this nding by proving that the RT distri-bution of reportedly intentional responses is bimodal, suggesting two different types of intentional responses: Early oneswhich were retrospectively constructed as intentional and late ones which actually were intentionally reinitiated. In the sec-ond Experiment we manipulated the amount of trials that were misattributed by varying the degree of deliberateness of theprimary response. In accordance with our predictions the amount of misattributed trials decreased when the primary re-sponse was intentionally initiated rather than externally induced. Previous research on intention ascription has demon-strated that participants sometimes have the illusion of conscious will when agency of the action is not clear. Ourndings demonstrate that such illusions of conscious will also occur when no other agent is involved.3.1.3. ProcedureThe design was similar to Experiment 1, with the exception that we used a different primary response task. If a dot ap-

peared on the left side of the screen the participants had to press the left button with the index nger, whereas if a dot ap-peared on the right side they had to press the right button with the middle nger of the right hand. These two stimuli arepart of what we call the stimulus condition, because the position of the dot indicates which response is required. When thedot appeared in the middle of the screen the participants had to choose between pressing the right or the left button. Wename this condition intention condition, because the subjects have to build up an intention which action to perform. Respec-tively on a scale of voluntariness of the primary response task the stimulus condition would score lower compared to theintention condition. Since responding to the position of a dot seems to be easier we chose to start with a staircase of250 ms (the staircase was separately modied for the stimulus vs. intention condition as well as for the stop vs. decide tri-als). Same as in Experiment 1 the dot stimuli changed colour after the staircase time elapsed. We used the same block struc-ture as in Experiment 1 with the difference that the subjects performed four sessions.

3.1.4. Data analysisPlease cite this article in press as: Khn, S.& Brass, M. Retrospective construction of the judgement of free choice. Conscious-ness and Cognition (2008), doi:10.1016/j.concog.2008.09.007


ARTICLE IN PRESS4.1. Signal detection on mental representations

Assuming that people more often envision success of a task than failure Wegners consistency principle states that thecorrespondence between the prior thought and the observed action leads to an experience of will when the success actuallyoccurs, but the experience of will should be markedly reduced, if the prior thought is incompatible to the action outcome. Inour experiments participants behave differently: they report an experience of will in a situation where they actually failed tosucceed; presumably because they did not detect this failure.

What people actually have to do when reporting on subjective states is to perform signal detection on their own mentalrepresentations, signal detection on the mind as Gaillard, Vandenberghe, Destrebecqz, and Cleeremans (2006) call it. Tosome extend this introspective signal detection seems to be correct, because participants are be able to discriminate casesin which they did decide (there were nearly no misses), but were not consistently capable of adequately judging when theyhave not been able decide (a lot of false alarms). There are trials in which the reaction times demonstrate that no stop anddecision process has taken place, but nevertheless participants indicate that they did make a voluntary choice to act. Onemight argue that our task is particularly confusing because participants have to prepare a response, have to inhibit that re-sponse, have to decide between acting and not acting, and nally have to indicate whether the act was voluntary or inducedby the target. Assuredly both actsresponses to letter stimuli and to dotsare voluntary as well, and might therefore beprone to be mixed-up with the free voluntary decision.

4.2. Condence about the awareness of intention

Although the task to determine the voluntariness of acts was not easy no participant expressed any uncertainty abouthow to discriminate between trials in which they performed a decision process and trials in which they did not. An indicatorshowing that the participants were highly condent about their deliberateness judgements is that they were faster to indi-

Fig. 6. (a) Lognormal tted reaction time distribution of decide-go trials in the stimulus condition of Experiment 2 (The y-axis represents a normalized scalesetting the area under the tted early & late decide-go curve to 1). (b) Lognormal tted reaction time distribution of decide-go trials in the intentioncondition of Experiment 2 (The y-axis represents a normalized scale setting the area under the tted early & late decide-go curve to 1).



ARTICLE IN PRESScate that they decided voluntarily after early trials that are supposedly retrospective constructions. Additionally participantsactually protested against not having been able to perfectly discriminate between a voluntary response and a default moderesponse when confronted with their data. Astonishingly at least one third of our subjects voiced nearly verbatim what Nisb-ett and Wilson (1977) report as experiences of social psychologists. Subjects typically said that the hypothesis was veryinteresting and that many people probably would go through the process that the experimenter described, but so far as theycould tell, they themselves had not. (p. 237). In our case the subjects pointed out, that other subjects might be lying but theythemselves answered faithfully during the whole experiment and knew very well in which trials they were able to decideand in which not. The great advantage of the experimental design we used is that we can recheck the introspective reportof the participants by consulting the associated RTs.

4.3. Two sources of introspection

Jack and Shallice (2001) describe introspective reports as the product of two factors, namely raw data, which partici-pants access via introspection and a model which the participants use to interpret the raw data. They claim that the crucialdifference between introspective and objective evidence is to be seen in the fact that objective evidence enables the re-searcher to go back to the raw data (see also Laming, 1997), which is exactly what we achieved with our modied stop task.Subsequently the authors propose introspection evidence as an adequate tool to search for tasks involving intentional action.According to their framework tasks involving intentional action recruit Type-C processes, whereas automatic actions donot. Those Type-C processes involve the supervisory attentional system (Norman & Shallice, 1986; Shallice, 1988) and aresaid to produce episodic encoding as a by-product. Applied to our experiment that would suggest that if the decision processis an intentional process it should involve those Type-C processes and should therefore require the accurate recollection ofthe presence of a decision process. But since participants mistake trials in which they have not been able to decide with trialsin which they have been able to decide reveals, following the reasoning of Jack and Shallice, that vetoing and decision makingis not an intentional act. We can conclude that subjects failed to detect the absence of both the stopping and deciding in theearly decide-go trials, because the knowledge of the absence of one of these processes would have been sufcient to inferthat they were not successful in the present trial. If subjects knew that they were not able to stop the prepared action,but pressed the button nevertheless, they should have been able to deduce that a decision was impossible. Therefore we con-clude that our participants wrongly attributed both processes: stopping and decision making to themselves in a considerableamount of decide-go trials. This clearly argues against Libets assumption (1999) that a veto process can be consciously ini-tiated. He used the veto in order to reintroduce the possibility to control the unconsciously initiated actions. But since thesubjects are not very accurate in observing when they have not stopped, the act of vetoing cannot be consciously initiated.

4.4. Who is the expert of judging intentions?

Wilson and Stone (1985) state that despite the vast amount of information people should have about themselves theirexplanations about their responses are no more accurate than the explanations of complete strangers. In his self-perceptiontheory Daryl Bem (1967) goes even further. He assumes that observations of our own behaviour are a major source of self-knowledge. The central proposition of his theory is that people infer their internal states just as an outside observer would,by seeing how they behave and guessing what intentions must cause that behaviour. Bem assumes that we analyze ourbehaviour in the same way an outside observer would: we look at our behaviour and make a guess about why we did it.

If one applies this idea of self-perception theory to our experiment, an illusion of will in which we experience action out-come as initiated by ourselves although it was actually produced by another agent and an illusion of will when no otheragent is involved is virtually the same. Resulting from that the ex-post evaluation of stopping and deciding might evenbe called confabulation after the fact.

But how comes that we nevertheless feel so condent that we do know when we stop an action and when we decide vol-untarily? According to a list of types of privately held knowledge by Jones and Nisbett (1972) one has to assert that thisknowledge outnumbers the knowledge of observers and it might become less surprising that people persist in believing thatthey have access to their own cognitive processes.

4.5. Why do we have an illusion of conscious will?

One explanation is that it might be important for people to feel that they are the well-informed captains of their ownship and know why they are doing what they are. As Wegner (2002, p. 217) puts it subjects are profoundly interestedin maintaining the ction that they have conscious will. Recognizing that we are no more informed about the causes ofour responses than a complete stranger would likely make people feel less in control of their lives, a feeling that has beenshown to be associated with depression (Benassi, Sweeney, & Dufour, 1988). On the contrary the illusion of conscious willseems to have positive effects concerning health as Taylor and Brown (1988) and Taylor (1989) describe. There are a lot ofinstances in which it seems to be better to think you have control, especially when suffering from serious illnesses.

To conclude, our ndings demonstrate that retrospective construction of the feeling of free choice can occur, presumablyespecially in cases when we are uncertain about the degree of deliberateness of an action. Further research is needed to ex-plore the precise prerequisites under which subjective back-referral emerges. The results do not rule out any predictivePlease cite this article in press as: Khn, S.& Brass, M. Retrospective construction of the judgement of free choice. Conscious-ness and Cognition (2008), doi:10.1016/j.concog.2008.09.007

mechanisms of motor control and do not imply that phenomenal awareness of action is in general a product of a reconstruc-tive mechanisms.

To summarize, our results suggest that we might sometimes know less than we think we do about our own minds. Butshould we tell you so? Freud might reply that the psychological defence against the idea that we do not know as much aboutourselves as we think we do will nevertheless keep us blind to the fact that any distortion is taking place. But if people knewthat they were changing their beliefs just to make themselves feel better, the change would probably not be as compelling.

Acknowledgements

The work was supported by the dissertation grant of the Studienstiftung des Deutschen Volkes awarded to the rst


ARTICLE IN PRESSauthor. We would like to thank Nils Bodammer for his patient MATLAB support.

References

Bays, P. M., Flanagan, R., & Wolpert, D. M. (2006). Attenuation of self-generated tactile sensations is predictive, not postdictive. PLoS Biology, 4(2), e28.Bem, D. J. (1967). Self-perception: An alternative interpretation of cognitive dissonance phenomena. Psychological Review, 74, 183200.Benassi, V. A., Sweeney, P. D., & Dufour, C. L. (1988). Is there a relationship between locus of control orientation and depression. Journal of Abnormal

Psychology, 97, 357366.Crowne, D. P., & Marlowe, D. (1960). A new scale of social desirability independent of psychopathology. Journal Consulting Psychology, 24, 349354.De Jong, R., Coles, M. G. H., Gratton, G., & Logan, G. D. (1990). In search of the point of no return: The control of response processes. Journal of Experimental

Psychology: Human Perception and Performance, 16(1), 164182.De Jong, R., Coles, M. G. H., & Logan, G. (1995). Strategies and mechanisms in nonselective and selective inhibitory motor control. Journal of Experimental

Psychology: Human Perception and Performance, 21(3), 498511.Dickenberger, D., Holtz, S., & Gniech, G. (1978). Bedrfnis nach sozialer Anerkennung: Validierung der Marlowe-Crowne Social Desirability Scale ber ein

Konzept individueller relevanter Gruppen. Diagnostica, 24(1), 2438.Eysenck, S. B. G., Daum, I., Schugens, M. M., & Diehl, J. M. (1990). A cross-cultural study of impulsiveness, venturesomeness and empathy: Germany and

England. Zeitschrift fr Differentielle und Diagnostische Psychologie, 11(4), 209213.Freud, S. (1917). Eine Schwierigkeit der Psychoanalyse. Gesammelte Werke Band (Vol. 12, pp. 11). Frankfurt am Main: Fischer Verlag.Gaillard, V., Vandenberghe, M., Destrebecqz, A., & Cleeremans, A. (2006). First- and third-person approaches in implicit learning research. Consciousness &

Cognition, 15, 709722.Aarts, H., Custers, R., & Wegner, D. M. (2005). On the inference of personal authorship: Enhancing experienced agency by priming effect information.

Consciousness & Cognition, 14, 439458.Jack, A. I., & Shallice, T. (2001). Introspective physicalism as an approach to the science of consciousness. Cognition, 79, 161196.Jones, E. E., & Nisbett, R. E. (1972). The actor and the observer: Divergent perceptions of the causes of behavior. In E. E. Jones et al. (Eds.), Attribution:

Perceiving the causes of behavior. Morristown, NJ: General Learning Press.Laming, D. R. J. (1997). The measurement of sensation, Oxford: Oxford University Press.Libet, B., Gleason, C. A., Wright, E. W., & Pearl, D. K. (1983). The time of conscious intention to act in relation to onset of cerebral activity (readiness-

potential): The unconscious initiation of a freely voluntary act. Brain, 106, 623642.Libet, B. (1999). Do we have free will? Journal of Consciousness Studies, 6(89), 4757.Logan, G. D. (1995). On the ability to inhibit thought and action: A users guide to the Stop Signal Paradigm. In D. Dagenbach & T. H. Carr (Eds.), Inhibitory

processes in attention, memory, and language. San Diego: Academic Press.Logan, G. D., & Cowan, W. B. (1984). On the ability to inhibit thought and action: A theory of an act of control. Psychological Review, 91(3), 295327.Musch, J., Brockhaus, R., & Brder, A. (2002). Ein Inventar zur Erfassung von zwei Faktoren sozialer Erwnschtheit. Diagnostica, 48(3), 121129.Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84(3), 231259.Norman, D. A., & Shallice, T. (1986). Attention to action: Willed and automatic control of behaviour. In R. J. Davidson, G. E. Schwarz, & D. Shapiro (Eds.).

Consciousness and self-regulation: Advances in research (Vol. 4, pp. 118). New York: Plenum Press.Oldeld, R. C. (1971). The assessment and analysis of handedness: The Edinburgh inventory. Neuropsychologia, 9(1), 97113.Paulhus, D. L. (1994). Balanced Inventory of desirable responding: Reference manual for BIDR version 6. Unpublished manuscript, University of British

Columbia, Vancouver, Canada.Preston, J., & Wegner, D. M. (2007). The Eureka error: Inadvertent plagiarism by misattributions of effort. Journal of Personality and Social Psychology, 92(4),

575584.Pronin, E., Wegner, D. M., McCarthy, K., & Rodriguez, S. (2006). Everyday magical powers: The role of apparent mental causation in the overestimation of

personal inuence. Journal of Personality and Social Psychology, 91(2), 218231.Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press.Taylor, S. E. (1989). Positive illusions: Creative self-deception and the healthy mind. New York: Basic Books.Taylor, S. E., & Brown, J. D. (1988). Illusion and well-being: A social psychological perspective on mental health. Psychological Bulletin, 193, 193210.Wegner, D. M. (2002). The illusion of conscious will. Cambridge: Bradford Book.Wegner, D. M., & Wheatley, T. P. (1999). Apparent mental causation: Sources of the experience of will. American Psychologist, 54, 480492.Wilson, T. D., & Stone, J. I. (1985). Limitations of self-knowledge: More on telling more than we can know. In P. Shaver (Ed.). Review of personality and social

psychology (Vol. 6, pp. 167183). Beverly Hills, California: Sage.Wolpert, D. M., Ghahramani, Z., & Jordan, M. I. (1995). An internal model for sensorimotor integration. Science, 269, 18801882.Please cite this article in press as: Khn, S.& Brass, M. Retrospective construction of the judgement of free choice. Conscious-ness and Cognition (2008), doi:10.1016/j.concog.2008.09.007

Retrospective construction of the judgement of free choiceIntroductionExperiment 1MethodsParticipantsApparatus and stimuliProcedureData analysisResults and discussion

Experiment 2MethodsParticipantsApparatus and stimuliProcedureData analysisResults and discussion

General discussionSignal detection on mental representationsConfidence about the awareness of intentionTwo sources of introspectionWho is the expert of judging intentions?Why do we have an illusion of conscious will?

AcknowledgementsReferences

kuhn 2008 retrospectivefreechoice

Documents