ORIGINAL RESEARCH
published: 29 Janua y 021
doi: 10.3389/ nhum.2021.610347
Edi ed by:
Je wen Jou,
The Uni e si y o Texas Rio G ande
Valley, Uni ed S a es
Re iewed by:
S e en Haase,
Shippensbu g Uni e si y,
Uni ed S a es
Ulla Ma ens,
Uni e si y o Osnab ück, Ge many
Yu-Cheng Lin,
The Uni e si y o Texas Rio G ande
Valley, Uni ed S a es
*Co espondence:
Lasse Güldene
[email p o ec ed]
Special y sec ion:
This a icle was submi ed
o Cogni i e Neu oscience,
a sec ion o he jou nal
F on ie s in Human Neu oscience
Recei ed: 25 Sep embe 2020
Accep ed: 04 Janua y 2021
Published: 29 Janua y 2021
Ci a ion:
Güldene L, Jüllig A, So o D and
Pollmann S (2021) Fea u e-Based
A en ional Weigh ing and
Re-weigh ing in he Absence
o Visual Awa eness.
F on . Hum. Neu osci. 15:610347.
doi: 10.3389/ nhum.2021.610347
Fea u e-Based A en ional Weigh ing
and Re-weigh ing in he Absence o
Visual Awa eness
Lasse Güldene 1*, An onia Jüllig1,Da id So o2and S e an Pollmann1,3,4
1Depa men o Expe imen al Psychology, O o- on-Gue icke-Uni e si y, Magdebu g, Ge many, 2Ike basque, Basque
Founda ion o Science, Basque Cen e on Cogni ion, B ain, and Language (BCBL), San Sebas ian, Spain, 3Depa men o
Expe imen al Psychology and Cen e o Beha io al B ain Science, O o- on-Gue icke-Uni e si y, Magdebu g, Ge many,
4Beijing Key Labo a o y o Lea ning and Cogni ion and School o Psychology, Capi al No mal Uni e si y, Beijing, China
Visual a en ion e ol ed as an adap i e mechanism allowing us o cope wi h a apidly
changing en i onmen . I enables he acili a ed p ocessing o ele an in o ma ion, o en
au oma ically and go e ned by implici mo i es. Howe e , despi e ecen ad ances
in unde s anding he ela ionship be ween consciousness and isual a en ion, he
unc ional scope o unconscious a en ional con ol is s ill unde deba e. He e, we
p esen a no el masking pa adigm in which olun ee s we e o dis inguish be ween
a ying o ien a ions o a b ie ly p esen ed, masked g a ing s imulus. Combining signal
de ec ion heo y and subjec i e measu es o awa eness, we show ha pe o mance on
unawa e ials was consis en wi h isual selec ion being weigh ed owa ds epea ed
o ien a ions o Gabo pa ches and ealloca ed in esponse o a no el unconsciously
p ocessed o ien a ion. This was pa icula ly p esen in ials in which he p io ea u e was
s ongly weigh ed and only i he no el ea u e was in isible. Thus, ou esul s p o ide
e idence ha in isible o ien a ion s imuli can igge he ealloca ion o his o y-guided
isual selec ion weigh s.
Keywo ds: ea u e-based a en ion, a en ional weigh ing, isual selec ion, cogni i e con ol, unconscious
INTRODUCTION
Fo su i al in an uns able and unce ain wo ld, i is c ucial o de ec con ex ual egula i ies, bu
also o adap quickly when hey change. Since such con ex ual changes may be complex and occu
e y apidly, he ques ion a ises as o whe he a en ion shi s in esponse o en i onmen al changes
a e con ingen on isual awa eness. P e ious s udies examined he e ec o exogenous in isible
cues on he deploymen o ex e nal isual selec i e a en ion, sugges ing ha subliminal spa ial
cues can cap u e a en ion and acili a e ask pe o mance a he cued loca ion (McCo mick, 1997;
Mulckhuyse e al., 2007; o a e iew see Mulckhuyse and Theeuwes, 2010), ha he associa ion
be ween a subliminal cue and a isible a ge can be lea ned implici ly (Lambe e al., 1999) and
ha subliminal s imulus can e en induce cogni i e con ol p ocesses like esponse inhibi ion o
ask-swi ching e ec s (Lau and Passingham, 2007; Van Gaal e al., 2008, 2010; Fa ooqui and Manly,
2015). This no ion is u he suppo ed by e idence om clinical s udies in ‘‘blindsigh ’’ pa ien s,
which indica e ha isual cues p esen ed in he pa ien ’s blind ield a e s ill capable o di ec ing
spa ial a en ion (Ken idge e al., 1999).
F on ie s in Human Neu oscience | www. on ie sin.o g 1Janua y 2021 | Volume 15 | A icle 610347
Güldene e al. Unconscious Re-weigh ing o Fea u e-Based A en ion
I is, howe e , less clea whe he ea u e-based a en ion can
be edi ec ed owa ds a no el ea u e ( ea u e-based a en ional
e-weigh ing) in esponse o changes in unconsciously p ocessed
a ge s: acco ding o Bundesen’s heo y o isual a en ion (TVA,
Bundesen, 1990), he a en ional selec ion is a mechanism ha
ope a es in he se ice o pe cep ual ca ego iza ion, i.e., by aiding
he selec ion o a po en ial a ge i em wi hin a dis ac o display
(‘‘ il e ing’’), o he disc imina ion o ea u es in single i ems
(selec ion o ca ego ies, ‘‘pigeonholing’’). The p ocessing speed
o his isual selec ion depends on bo h he a en ional weigh
and he pe cep ual decision bias. In heo y, he a en ional weigh
elies on he senso y e idence indica ing he ca ego y a ce ain
s imulus belongs o (‘‘bo om-up’’), and he goal- ele ance o ha
ca ego y i.e. he impo ance o a ending o a ce ain s imulus
ca ego y (‘‘ op-down’’; Bundesen, 1990). Thus, he weake he
senso y e idence is, he mo e he a en ional weigh ing should
ely on he ‘‘ op-down’’ mechanism ( he impo ance o a end
o his ca ego y). Based on he TVA’s assump ion a ‘‘ op-
down’’ d i en a en ional bias (i.e., he goal- ele ance) on
selec ion is p edic ed especially o in isible non-consciously
p ocessed isual s imuli because he senso y e idence ha
could suppo isual selec ion in a bo om-up ashion (i.e., he
saliency o he s imulus) is e y limi ed i he s imulus is only
unconsciously pe cei ed. Impo an ly, e idence is s ill missing
as o whe he such a ea u e-based selec ion bias can be elici ed
o subliminal, unconsciously p ocessed s imuli and whe he i
can be eweigh ed lexibly in esponse o ea u e changes o he
unconsciously p ocessed s imulus.
La e accoun s o isual a en ion c i icize he dicho omy o
bo om-up s. op-down a en ional weigh ing and p opose o
include a his o y-d i en weigh ing o a en ional selec ion (e.g.,
Awh e al., 2012; Theeuwes, 2018, 2019) o be e inco po a e
empi ical e idence showing ha no only can s imulus saliency
and in e nal goals ( oli ional con ol) bias a en ional selec ion
bu he ‘‘his o y’’ o o me a en ion deploymen s d i en by
e.g., ewa d, in e ial p iming, o s a is ical lea ning (Awh e al.,
2012) can also ha e an in luence. Fo consciously pe cei ed
isual s imuli, such his o y-d i en a en ion weigh ing e ec s
ha e been obse ed in single on sea ch asks. Fo ins ance,
epea ed p esen a ion o he same a ge -de ining dimension
leads o esponse ime bene i s and associa ed ac i a ion changes
in dimension-speci ic isual p ocessing a eas (Pollmann e al.,
2006) ha we e in e p e ed as e idence o an a en ional
weigh ing o he a ge -de ining dimension (Mülle e al., 1995;
Liese eld e al., 2018). In con as , when he a ge -de ining
dimension changes, e.g., when he a ge was de ined by a
single on colo in ecen ials and hen is de ined by a single on
mo ion di ec ion, esponse ime cos s a e obse ed, as would
be expec ed when a en ion needs o be eweigh ed o he
new a ge -de ining dimension. These eweigh ing p ocesses
occu inciden ally, in he absence o an explici ins uc ion
o a end o he new a ge -de ining dimension (Mülle e al.,
2004). Fu he mo e, a compa able spa ial a en ion weigh ing
pa e n is obse ed when implici ly lea ned a ge -dis ac o
con igu a ions change in he con ex ual cueing pa adigm
(Manginelli and Pollmann, 2009; Pollmann and Manginelli,
2009). When a en ion-weigh ing p ocesses occu in he absence
o explici ask demand and e en a e changes o implici ly
lea ned con igu a ions, he nex ques ion would be whe he
a en ional eweigh ing can also occu as an adap i e adjus men
o unconsciously pe cei ed s imulus changes.
The e o e, his s udy add essed wo key ques ions. Fi s , we
asked whe he he epea ed p esen a ion o an in isible a ge
ea u e can lead o a empo ally pe sis ing a en ional selec ion
bias. The second ques ion was how lexible his a en ional bias
is, i.e., whe he a no el in isible a ge can igge he eweigh ing
o isual a en ion o he new a ge ea u e in he absence o
awa eness. Pe emen e al. (2013) s udied he ela ion o in e ial
ea u e p iming and isual awa eness du ing a le e sea ch ask.
They epo ed ha he epe i ion o he a ge shape speeded
isual sea ch only when he a ge in he p ime display had
been consciously pe cei ed. Ye , i emains unknown whe he
unconscious eweigh ing o isual selec ion can occu o simple
o ien a ion s imuli such as Gabo pa ches (Rajimeh , 2004). We
also conside ed a di e en ask se ing in which he selec ion ask
occu ed a a ixed a ended loca ion h oughou he ials. In all
p e ious s udies, a en ion-weigh ing e ec s we e examined in
mul i-i em displays and sea ch asks o a single on a ge . Ou
pa adigm does no in ol e spa ial shi s o a en ion bu a he a
p ocess o isual selec ion in which he same spa ial loca ion is
always a ended.
Speci ically, ou pa adigm in ol ed an o ien a ion
disc imina ion ask based on a cen al masked ba s imulus.
Volun ee s we e ins uc ed o disc imina e whe he he a ge
s imulus was e ical o il ed i espec i e o he speci ic di ec ion
o il . They had o make no u he dis inc ion be ween he wo
il ed o ien a ions. Ye , o in oduce he il -based a en ional
selec ion bias, we manipula ed he likelihood o he wo
non- e ical g a ings (le s. igh ) so ha one il would occu
wice as o en as he o he . Consis en wi h he p opo ion
cong uency e ec du ing p iming (Bodne and Lee, 2014; Blais
e al., 2016), and ea u e-based s a is ical lea ning (Tu k-B owne
e al., 2009; Che e iko e al., 2017), an inc ease o he equency
a which a igh o le - il ed g a ing appea ed should esul in
a high selec ion weigh o he equen o ien a ion indica ing
he impo ance o a end o his ca ego y. This p edic ion is
based on he idea ha he ele an ea u e in o ma ion (e.g., he
spa ial o ien a ion) o he mos likely a ge ge s ep esen ed in
a o m o a sho - e m desc ip ion— he a en ional empla e
(Desimone and Duncan, 1995), o con ol he senso y p ocessing
so ha s imuli ma ching he desc ip ion a e a o ed, i.e., a e
mo e eadily p ocessed in he isual sys em. The deg ee o
which a s imulus ma ches he a en ional empla e de ines i s
a en ional weigh . Thus, Gabo pa ches ha i he in o ma ion
s o ed in he empla e ecei e a high selec ion weigh , e.g., 1,
while misma ching Gabo pa ches (in equen and e ical)
ha e educed selec ion weigh s as he whole weigh is hough
o be a cons an alue: i he weigh inc eases o one ea u e
i dec eases o ano he (Duncan and Humph eys, 1989).
Now, conce ning beha io a swi ch om he hea ily weigh ed
o ien a ion o a a ge wi h a e ical o he in equen spa ial
o ien a ion should equi e a shi o selec ion weigh s due o
he misma ch be ween he senso y inpu and he a en ional
empla e. This shi o a en ional selec ion weigh s was expec ed
F on ie s in Human Neu oscience | www. on ie sin.o g 2Janua y 2021 | Volume 15 | A icle 610347
Güldene e al. Unconscious Re-weigh ing o Fea u e-Based A en ion
o lead o slowing s imulus p ocessing and esponse ini ia ion
e en ually esul ing in inc eased esponse la encies in such
swi ch ials. The highe he selec ion bias o he Gabo pa ch’s
o ien a ion in he p eceding ial, he mo e eweigh ing should
be necessa y o p ocess and espond o a no el g a ing in he
subsequen ial. Thus, pa icula ly swi ch ials in which he
p io o ien a ion was he highly equen il should show
p olonged esponse la encies on he beha io al le el, gi en ha
he inc eased likelihood o one o ien a ion o e he o he s was
su icien o induce a p io selec ion bias (e.g., Lebe e al., 2009;
Che e iko e al., 2017). Impo an ly, a combina ion o signal
de ec ion heo e ic measu es (S anislaw and Todo o , 1999) and
subjec i e pe cep ual a ings (Ramsøy and O e gaa d, 2004)
was used o assess pa icipan s’ awa eness o he s imulus o
a oid po en ial con ounds due o c i e ion biases in epo ing
(un)awa eness, e.g., epo s o no expe ience o he knowledge
held wi h low con idence (Wiens, 2007; So o e al., 2019).
The e o e he unconscious eweigh ing o selec ion hypo hesis
was e en ually es ed by main aining a clea sepa a ion be ween
he measu es o selec i e a en ion weigh ing, in e ed by he
pa e n o esponse la encies, and he measu es ha we used
o p obe (un)awa eness o he s imulus (objec i e o ien a ion
disc imina ion ask and subjec i e epo s). We p edic ed
decision eac ion ime (RT) cos s due o a change o he il
di ec ion. Cos s should be highes i he p io o ien a ion was
he highly biased il , i.e., a swi ch om he equen o he
in equen il o a e ical a ge , and hey should occu e en i
he no el a ge is non-consciously pe cei ed.
MATERIALS AND METHODS
Pa icipan s
In o al 21 na i e Ge man s uden s ( h ee male) om
he Uni e si y o Magdebu g, Ge many ook pa in he
expe imen . All olun ee s we e be ween 19 and 34 yea s old
(M= 24.90 yea s), igh -handed by sel - epo excep o one
pa icipan , and had a no mal o co ec ed- o-no mal ision.
They p o ided w i en consen and we e ei he mone a ily
eimbu sed (8 eu os pe hou ) o ecei ed cou se c edi s o he
2 h o pa icipa ion. In wo sessions an e o in he esponse
collec ion occu ed and he espec i e pa icipan s we e emo ed
om he analysis. Ano he olun ee in e up ed he session a
an ea ly s age and was hus excluded. Du ing da a analysis, i e
o he pa icipan s we e iden i ied o ha e mo e han 40% missing
esponses du ing he 1.5 s esponse deadline (see below) and we e
hus excluded om RT analysis.
Appa a us and S imuli
The s imulus display and esponses we e con olled wi h
he Py hon oolbox ‘‘Psychopy’’ (Pei ce, 2007; Pei ce e al.,
2019). The s imuli we e p esen ed on a 2400 Samsung moni o
(1,920:1,080 esolu ion, 60 Hz e esh a e). All pa icipan s we e
placed 50 cm away om he sc een. S imuli we e Gabo g a ings
wi h an indi idually calib a ed con as (see ‘‘Expe imen al
Task and P ocedu es’’ sec ion) cen ally p esen ed on a g ay
backg ound sub ending 3.4◦ isual angle. I s spa ial equency
was 3.7◦cycles pe deg ee. The pa ch’s o ien a ion was ei he
e ical (180◦), 165◦, 150◦, o 135◦i i was a le - il ed,
non- e ical Gabo pa ch, and 195◦, 210◦, o 225◦i i was
a non- e ical pa ch il ed o he igh . To u he educe he
isibili y o he Gabo pa ch we used a ci cula backwa d mask
o black and whi e andom do s (3.4◦ isual angle).
Expe imen al Task and P ocedu es
Th eshold De e mina ion
A session s a ed wi h a s ai case p ocedu e o calib a e he
s imulus’s luminance con as ende ing i s o ien a ion in isible.
Gabo pa ches occu ed cen ally on he sc een o 33.33 ms
(i.e., he g a ing was p esen ed o wo ames, each o which had
a minimal p esen a ion du a ion o 1/60 ms) and we e di ec ly
ollowed by he mask o 350 ms. I pa icipan s saw he g a ing’s
o ien a ions, hey we e o espond by p essing he ‘‘up’’-key,
while he ‘‘down’’-key was o be p essed i hey did no see he
o ien a ion. In he main expe imen olun ee s we e o a e hei
subjec i e isibili y o he a ge a he end o each ial using
he ou -poin pe cep ual awa eness scale (PAS): (1) ‘‘did no
see any hing a all,’’ (2) ‘‘saw a b ie glimpse wi hou seeing he
o ien a ion,’’ (3) ‘‘had an almos clea image o he s imulus,’’
and (4) ‘‘saw he s imulus and i s o ien a ion’’ (Ramsøy and
O e gaa d, 2004). Du ing ini ial calib a ion, pa icipan s we e
hus ins uc ed o epo no expe ience o he s imulus only
i hey did no see any hing a all which co esponded o he
i s poin o he PAS. Con e sely, hey we e asked o indica e
an awa e esponse in ials whe e a b ie glimpse o a mo e
s able pe cep o he Gabo was expe ienced co esponding
o he emaining h ee poin s o he PAS. The s imulus’
luminance con as was dec eased ollowing an awa e esponse
and inc eased ollowing an unawa e esponse. All pa icipan s
did 90 ials (30 ials o each o he h ee o ien a ions). The inal
h eshold luminance was de ined as he mean luminance con as
ac oss he las 10 ials o he s ai case.
Nex , pa icipan s pe o med one block o aining unde
expe imen al condi ions consis ing o 36 p ac ice ials. He e he
luminance con as ob ained a e he i s s ai case p ocedu e
was used o he con as alue o he aining s imuli. The
p ac ice uni was ollowed by a second calib a ion conduc ed
acco ding o he same p o ocol as he i s s ai case p ocedu e.
E en ually, he second ecalib a ion p o ided he h eshold alue
o he luminance con as used in he main ask.
Task
In he main expe imen , olun ee s we e asked o pe o m an
o ien a ion disc imina ion ask based on masked Gabo pa ches.
The s a o a new ial was signaled by a cen al ixa ion
emaining 500 ms on he display ollowed by a blank sc een
o ano he 500 ms du a ion. Then he a ge Gabo pa ch
occu ed a he sc een cen e o 33.33 ms. A pa e n backwa d
mask (B ei meye and Ogmen, 2000) ollowed immedia ely o
350 ms. In he ollowing 1.5 s pa icipan s we e o gi e hei
disc imina ion esponse. A he ial’s end, hey we e e en ually
p omp ed o a e he isibili y o he Gabo pa ch using he
keys 1–4 wi hin he nex wo 2 s. All ials we e sepa a ed
by in e - ial-in e als (ITI) wi h a ying du a ions (1.5–3.5 s)
F on ie s in Human Neu oscience | www. on ie sin.o g 3Janua y 2021 | Volume 15 | A icle 610347
Güldene e al. Unconscious Re-weigh ing o Fea u e-Based A en ion
FIGURE 1 | Example o a ial sequence. The box a he op shows an example o he epea condi ion (le ): a e ical a ge g a ing in he i s ial is ollowed by
ano he e ical g a ing in he second ial. On he igh i shows an example o he swi ch condi ion: he le - il ed a ge g a ing is ollowed by a igh - il ed g a ing in
he nex ial.
ollowing a loga i hmic dis ibu ion. Figu e 1 depic s a de ailed
ial sequence.
Design
To acili a e he occu ence o a il -based a en ional selec ion
bias, we in oduced une en p opo ions o he wo non- e ical
g a ings (le s. igh ). Consis en wi h ea u e-based s a is ical
lea ning (e.g., Tu k-B owne e al., 2009; Che e iko e al.,
2017), he ela i e inc ease o he equency a which a igh
o le - il ed g a ing appea ed was expec ed o s ongly weigh
a en ional selec ion o his o ien a ion. I s highe likelihood
should inc ease he impo ance o a ending o his ea u e,
esul ing in a high selec ion weigh . A he same ime, he
selec ion weigh o he o he wo o ien a ions ( e ical and he
in equen il ) should be educed (Bundesen, 1990). E en ually,
swi ches away om he hea ily weigh ed il we e expec ed o
esul in a signi ican inc ease in olun ee s’ esponse imes.
The e o e, o a block o 36 ials, we chose 12 e ical a ge s
(∼33%), and used une en p opo ions o he wo il s, wi h
18 ials (50%) and six ials (∼16%), espec i ely. This way, each
block was ei he le —(75% o all non- e ical ials we e le -
il ed) o igh —weigh ed (75% o all il ials we e igh - il ed).
The ac ual p esen a ion o he h ee o ien a ions was andomized
wi hin a single block.
The i s 11 pa icipan s pe o med 14 blocks in he
main expe imen (504 ials). The eigh h subjec , howe e ,
in e up ed he session a e 12 blocks we e comple ed. Subjec s
12–21 comple ed 10 blocks (360 ials) as his amoun o ials
u ned ou o be su icien o ob ain enough ials o each
awa eness le el (AL) while a oiding g owing wea iness ha was
epo ed by subjec s comple ing 14 blocks.
S a is ical Analysis
Sensi i i y and esponse bias measu es we e calcula ed using
cus om-made Py hon code (Ve sion 2.7). All s a is ical analyses
we e ca ied ou wi h R (Ve sion 3.5, R Co e Team, 2014). Fo
he Bayes ac o (BF) analysis (Roude e al., 2009) we used JASP
(JASP Team, 2019).
F on ie s in Human Neu oscience | www. on ie sin.o g 4Janua y 2021 | Volume 15 | A icle 610347
Güldene e al. Unconscious Re-weigh ing o Fea u e-Based A en ion
Subjec i e Awa eness
Fo each pa icipan , he numbe o ials o each subjec i e AL
was coun ed using he ial-by- ial PAS- a ing.
Disc imina ion Pe o mance
To examine whe he he pa icipan ’s abili y o co ec ly
disc imina e be ween e ical and non- e ical g a ings depends
on he le el o subjec i e awa eness, we i s ly de e mined
indi idual esponse bias and pe cep ual sensi i i ies using signal
de ec ion heo y (S anislaw and Todo o , 1999; Macmillan and
C eelman, 2004). G oup-e ec s we e subsequen ly assessed o
each le el o subjec i e awa eness using BFs since i was equi ed
o p o e he absence o sensi i i y (H0; Gallis el, 2009; Dienes
and Mcla chie, 2018). A BF(10)p o ides mode a e e idence
o H0 (e.g., A’ = 0.5) i i s ands be ween 0 and 0.33,
anecdo al e idence i i s ands be ween 1/3 and 1, and e idence
o H1 (A’ >0.5) i i exceeds 1 (Dienes and Mcla chie,
2018), wi h a BF(10)be ween 1 and 3, 3 and 10, 10 and
30, 30 and 100 and >100 p o iding anecdo al, mode a e,
s ong, e y s ong, and ex eme e idence, espec i ely, o H1
(Je eys, 1998; Quin ana and Williams, 2018).
Unde Yes/No-condi ions A’ and he c i e ion loca ion (C)
we e calcula ed o de e mine pe cep ual sensi i i y and bias:
we calcula ed alse-posi i e a es [FPR = False ala ms/(False
Ala ms + Co ec Rejec ions)] and hi a es [TPR = Hi s/(Hi s
+ Misses)] de ining a hi as he co ec epo o a non- e ical
o ien a ion when he Gabo ’s o ien a ion uly was il ed; alse
ala ms we e de ined as il esponse o e ical g a ings. We used
he ollowing o mulas o calcula e he non-pa ame ic esponse
bias and sensi i i y (S anislaw and Todo o , 1999):
C= −[Z(TPR)+Z(FPR)]/2
A0=0.5 + |sign(TPR −FPR;(TPR −FPR)2+
|TPR −FPR|/(4max(TPR,FPR)−4∗TPR∗FPR))|
Values o C a ound 0 indica e unbiased disc imina ion
pe o mance. A libe al decision c i e ion a o ing yes- esponses
(i.e., epo ing a non- e ical g a ing) leads o alues o C<0,
while posi i e alues occu i pa icipan s a e biased o epo
a e ical a ge . I olun ee s possess pe ec sensi i i y a
disc imina ing he a ge o ien a ions, A’ appea s o be equal o
1 and i dec eases o 0.5 i he sensi i i y diminishes (S anislaw
and Todo o , 1999).
Analysis o RT Da a
We used he packages lme4 (Ba es e al., 2015) as well as lme Tes
o make use o a linea mixed model (LMM) analysis. As he da a
was unbalanced due o he a ia ions in he subjec i e awa eness
a ings (PAS) ha lead o une en numbe s o ials ac oss he
ou le els o isual awa eness, LMMs we e chosen o e cus om
epea ed measu es ANOVAs o analyze he RT (e.g., A neon
and Lamy, 2018). Since all cases wi h missing da a would be
excluded in a epea ed-measu es ANOVA, he LMM app oach is
he be e means o make use o all a ailable da a in he ace o an
unbalanced design (Magezi, 2015). Only RTs o ials wi h co ec
esponses en e ed he analysis a e each pa icipan ’s indi idual
ou lie s (mean RTs ±3SD) we e emo ed.
Be o e assessing he signi icance o he ixed e ec s, we
de e mined he andom e ec s uc u e o he inal model wi h
likelihood a io es s (i.e., compa isons o models di e ing in
hei andom e ec s uc u e). Impo an ly, we did no use
likelihood a io es s o compa e models wi h di e ences in
hei ixed e ec s as hese we e al eady de e mined by he
design (see below). Once he inal model o analysis was ully
de ined, we i ed his model wi h he RT da a using a es ic ed
maximum likelihood es ima ion (REML) and es ed he s a is ical
signi icance o he ixed e ec p edic o s wi h a ype III ANOVA
wi h F-s a is ics as implemen ed in he lme unc ion o he
lme4 package (Ve sion 1.1–23; Richa dson and Welsh, 1995;
Bolke e al., 2009; Luke, 2017; McNeish, 2017). The p- alues
we e calcula ed using Sa e hwai e app oxima ions o deg ees
o eedom wi h he ANOVA unc ion o he package lme Tes
(Ve sion 3.1-2, Kuzne so a e al., 2017). We chose he ANOVA
app oach o es he s a is ical signi icance o he ixed e ec s
as his app oxima ion is hough o be p oducing accep able
Type 1 e o a es e en o small samples while he use o
model compa isons (likelihood a io es s) is no ecommended
o es ixed e ec s because hey appea o be an i-conse a i e
(Pinhei o and Ba es, 2000;Bolke e al., 2009; Luke, 2017). Pos
hoc es s (leas squa ed means o he con as s wi h Bon e oni
co ec ion) we e pe o med using he R package emmeans
(Ve sion 1.4.7). Finally, we used he R unc ion .squa edGLMM
as implemen ed in he R package MuMin o calcula e he
ma ginal Rsqua ed (R2
m) and condi ional Rsqua ed (R2
c) o
ob ain s anda dized e ec sizes. R2
mis in e p e ed as he a iance
explained by he ixed e ec s o awa eness and swi ch and R2
c
gi es he a iance explained by all ixed and andom e ec s
(Johnson, 2014).
The main goal we pu sued in he s udy was he examina ion
o whe he a changing o ien a ion om one ial o ano he
(swi ch) a ec ed pa icipan s’ esponses: we p edic ed a swi ch-
ela ed slowing o RTs compa ed o ials in which he o ien a ion
emained unchanged ( epea ). Thus, he swi ch o o ien a ions
(swi ch s. epea ) cons i u ed he i s ixed e ec p edic o in
he LMM. RTs we e also expec ed o dec ease wi h inc easing
isual awa eness: he mo e he pa icipan s saw, and he mo e
con iden ly hey should pe o m a ca ego izing he s imulus
o ien a ion, he as e hey should be a esponding o he
g a ing’s o ien a ion. The e o e, isual awa eness was de ined
as he second ixed e ec p edic o o he basic model. Finally,
o make allowance o a possible in e ac ion be ween he
wo ixed e ec s we included he in e ac ion e m o swi ch
and awa eness in o he inal LMM. Rega ding in e indi idual
baseline di e ences in esponse la encies, we also de ined a
by-subjec andom in e cep accoun ing o non-independency
o single subjec s’ da a. Thus, he basic model was o malized as
RT ∼swi ch + awa eness + swi ch:awa eness (1 | subjec ).
In his model, howe e , he ull andom e ec s uc u e
s ill needed o be de e mined. The e o e, we nex used model
compa isons based on likelihood a io es s (χ2) wi h he
ANOVA unc ion o he lme4 package (Baayen e al., 2008)
o assign he ull andom e ec s uc u e (Ba e al., 2013)
o his basic model. De ining he andom e ec s uc u e is
impo an o balance be ween he ype I e o a e ha in la es
F on ie s in Human Neu oscience | www. on ie sin.o g 5Janua y 2021 | Volume 15 | A icle 610347
Güldene e al. Unconscious Re-weigh ing o Fea u e-Based A en ion
i he andom e ec s uc u e o an LMM is unde speci ied
(Ba e al., 2013), and he model powe ha su e s i he
andom e ec s uc u e is mo e complex han he gi en da a
(Ma uschek e al., 2017). The me hod o model compa isons
based on likelihood a io es s compa es o he p ocedu e
o a hie a chical eg ession in which ele an p edic o s a e
added o he eg ession model and kep i hey signi ican ly
imp o e he model i (changes in R2). Likelihood a io
es s a e deemed o be app op ia e o o mally de ine he
andom e ec s uc u e o an LMM e en i he sample size
is small (Baayen e al., 2008; Bolke e al., 2009). Using his
me hod, we es ed he basic model con aining only a by-subjec
in e cep agains al e na i e models con aining an addi ional
by-subjec andom slope o awa eness and/o a by-subjec
andom slope o he swi ch. The de ails o his analysis
a e epo ed in he Supplemen a y Ma e ial. Impo an ly, we
used he likelihood a io es s only o de e mine he andom
e ec s uc u e o he inal model ha we used o i he RT
da a wi h, while he signi icance o he ixed e ec s (i.e., he
hypo heses es ing) was assessed using he ype III ANOVA
wi h Sa e hwai e app oxima ions o deg ees o eedom (Luke,
2017). Based on he model compa isons we included a by-subjec
andom slope o awa eness o model po en ial by-subjec
he e oscedas ici y conce ning awa eness [i.e., allowing une en
a iances ac oss he le els o he ixed e ec awa eness (Baayen
e al., 2008)]. E en ually, he inal model o signi icance es ing
was de ined as RT ∼swi ch + awa eness + swi ch:awa eness (1 +
awa eness | subjec ).
The inal LMM wi h he s uc u e ou lined abo e was applied
in wo RT models: In he i s model (a e age RT model) we
included all possible o ien a ion changes in he swi ch condi ion.
In he second model (weigh ed RT model) he swi ch condi ion
con ained only hose swi ch ials in which we expec ed he
highes RT cos s o occu : he equency di e ences be ween he
h ee o ien a ions we e expec ed o boos he selec ion weigh
o he highly equen non- e ical o ien a ion (ei he le o
igh ). Consequen ly, e-weigh ing o he in equen non- e ical
o ien a ion should be associa ed wi h mo e p onounced swi ch
cos s han ice e sa. The same was p edic ed o changes away
om he hea ily weigh ed o he e ical o ien a ion equi ing
s onge a en ional e-weigh ing. Howe e , swi ches away om
he low- equen il ed o ien a ion o e ical should lead o less
p ominen RT cos s because he a en ional selec ion weigh o
his il ed o ien a ion is weake , acili a ing he shi o a en ional
esou ces owa ds he no el a ge o ien a ion. Hence, hese ials
we e no included in he weigh ed RT model. We sepa a ely
epo he esul s o he LMM analyses o he a e age and he
weigh ed RT model.
RESULTS
Subjec i e Awa eness
In he majo i y o ials, pa icipan s’ subjec i e awa eness o
he o-be-disc imina ed o ien a ion was low (AL2, 25.94%), o
epo ed expe ience was ully absen (AL1, 37.70%). In abou
26.80% o all ials, subjec s epo ed an almos clea pe cep ion
o he g a ing (AL3) and i s o ien a ion. In only 9.54% o all ials
did hey clea ly see he g a ing and i s o ien a ion (AL4). Due o
he low numbe o hese AL4 ials, we excluded hem om he
ollowing analyses. A de ailed summa y o he numbe o ials
o swi ch and epea ials o each le el o subjec i e awa eness
is epo ed in he Supplemen a y Ma e ial.
A e he expe imen al session, each olun ee was asked o
epo whe he any di e ences in he equencies o he s imulus
o ien a ion had been no iced. The majo i y o subjec s epo ed
ha mo e il ed han e ical o ien a ions had been p esen ed
bu none o he pa icipan s no iced he block-wise changing
equency di e ence o he il ed o ien a ions (le s. igh ).
Objec i e Disc imina ion Abili y and
Subjec i e Awa eness Conco dan ly
Diminish
Signal de ec ion analyses e ealed ha on ials wi h almos
ull (AL3) and pa ial awa eness (AL2) pa icipan s’ pe cep ual
sensi i i y was signi ican ly abo e chance. Bayes- ac o s p o ided
ex eme e idence ha sensi i i y (A’) was g ea e han 0.5 in AL3,
BF(10)>100, 95% CI (0.733, 0.845), and AL2 ials, BF(10)>100,
95% CI (0.672, 0.750) (Quin ana and Williams, 2018). The mean
A’ o 0.789 ±0.026 (SE) in AL3 ials was 7.6 imes mo e likely
o be g ea e han he mean A’ o 0.711 ±0.018 in AL2 ials,
BF(10)AL3 >AL2 = 7.608 (Quin ana and Williams, 2018).
On unawa e ials (AL1), howe e , we obse ed a mean A’ o
0.516 ±0.030 ha was mo e likely o be equal o 0.5 wi h
mode a e e idence o he H0, BF(10)= 0.317, 95% CI (0.451,
0.582) (Quin ana and Williams, 2018), indica ing he absence o
pe cep ual disc iminabili y o he g a ings’ o ien a ion.
In con as o he pe cep ual sensi i i y analyses, indi idual
esponse biases emained una ec ed by changes in subjec i e
awa eness. In none o he ou ALs did we ind clea e idence o
a mo e libe al o mo e conse a i e esponse c i e ion o epo
a non- e ical o ien a ion, han a C a ound 0. BFs we e a he
in a o o he null hypo hesis indica ing ha he mean C o
−0.188 ±0.173 in AL1 ials was mo e likely no di e en om
ze o, ye wi h only anecdo al e idence o he H0, BF(10)= 0.457,
95% CI (−0.560, 0.188) (Quin ana and Williams, 2018). The
same was ue o he mean C o −0.073 ±0.106 in AL2 ials,
BF(10)= 0.342, 95% CI (−0.303, 0.157), and o a mean C o
0.089 ±0.123 in AL3 ials, BF(10)= 0.349, 95% CI (−0.179,
0.357) (Quin ana and Williams, 2018).
To examine a ia ions in he esponse c i e ion loca ion
(C) ac oss he h ee le els o subjec i e isual awa eness, we
made use o LMM o op imally deal wi h he unbalanced da a
se (Magezi, 2015). Since we aimed o assess he absence o
a ia ions in C ac oss he h ee le els o subjec i e awa eness, we
conduc ed a Bayesian-based LMM analysis using he R package
BayesFac o o ob ain a Bayes ac o (BF(10)) di ec ly p o ing he
null hypo hesis (Mo ey and Roude , 2018): i s we cons uc ed
a null model in which only a by-subjec andom in e cep was
included assuming ha a ia ions in C elied on in e indi idual
di e ences only [C ∼0 + (1| subjec )]. Nex , we cons uc ed
an al e na i e model in which he subjec i e awa eness epo s
(awa eness a ings 1–3) se ed as a single ixed e ec explaining
a iance in C in addi ion o he by-subjec andom in e cep [C
F on ie s in Human Neu oscience | www. on ie sin.o g 6Janua y 2021 | Volume 15 | A icle 610347
Güldene e al. Unconscious Re-weigh ing o Fea u e-Based A en ion
∼awa eness + (1 | subjec )]. Using he lmBF unc ion we hen
calcula ed BFs o each model and compa ed he wo models
by di iding he BF o he model ha included awa eness as a
ixed e ec by he BF o he null model. The analysis esul ed
in an inconclusi e BF(10)o 0.58 p o iding weak e idence o
he absence o a ia ion in C ac oss he h ee le els o subjec i e
awa eness (Quin ana and Williams, 2018).
Desc ip i e da a o sensi i i y and bias measu es a e epo ed
in Table 1. The da a showed ha he pe o mance o a leas wo
subjec s was highly biased in ials a ed as ully unawa e wi h a
shi in C o +1 SD and −1SD, espec i ely. A g aphic illus a ion
o he ela ion be ween he objec i e measu es o awa eness and
he subjec i e measu e is shown in Figu e 2 depic ing iolin
plo s o A’ and C o each le el o subjec i e awa eness. In sum,
hese esul s show ha he abili y o dis inguish he wo ypes o
o ien a ion (non- e ical s. e ical) s ongly depended on he
subjec i e isibili y and ully diminished on subjec i ely unawa e
ials. In con as , he e was no clea e idence o a esponse bias,
ega dless o he le el o subjec i e awa eness. Impo an ly he
absence o a ia ion in olun ee s’ esponse bias likely sugges ed
ha pe cep ual decision c i e ia we e no dependen on he
awa eness epo s and ha a ia ions in pa icipan s’ pe cep ual
sensi i i y ega ding he s imulus o ien a ion could hus no be
caused by a ia ions in he esponse bias.
In he signal de ec ion analysis ou lined abo e, we assumed
a bina y yes/no ask se up. Howe e , as we deployed le and
igh - il ed g a ings nex o e ical ones, subjec s we e o so
h ee possible s imulus ypes in o wo ca ego ies. Mo eo e ,
we p esen ed le - and igh - il ed Gabo s wi h a ying angles
so ha subjec s needed o map di e en s imuli o he same
esponse. Hence, a classi ica ion scena io may be e i he
scena io (Snodg ass e al., 2004). Impo an ly, such a se up
equi es he implemen a ion o wo a he han one decision
c i e ion inc easing he decision unce ain y, and he p opo ion
o co ec esponses [i.e., p opo ion co ec ,p(c)] is hen used
o measu e olun ee s’ classi ica ion sensi i i y (Macmillan and
C eelman, 2004, p. 190–191). Hence, ou sensi i i y measu e may
no be exhaus i e o all he in o ma ion ha he subjec could
hold, meaning ha ac ual sensi i i y on unawa e ials could be
highe han we measu ed.
Thus, we addi ionally calcula ed p(c) o each le el o
subjec i e awa eness (AL1–AL3): p(c) can be de ined as he
p io p obabili y o a posi i e s imulus (i.e., non- e ical g a ing)
imes he condi ional p obabili y o a posi i e esponse gi en a
posi i e s imulus (i.e., a non- e ical esponse o a non- e ical
a ge ) added o he p oduc o he p io p obabili y o he
nega i e s imulus (i.e., e ical) imes he condi ional p obabili y
o a nega i e esponse gi en a nega i e s imulus (Swe s, 2014,
TABLE 1 | A e age sensi i i y (A’) and c i e ion loca ion (bias, C) o he
o ien a ion disc imina ion ask o each le el o subjec i e awa eness.
Awa eness
Le el 1 Le el 2 Le el 3
A’ C A’ C A’ C
M 0.515 0.048 0.670 −0.093 0.723 −0.152
SD 0.091 0.438 0.110 0.402 0.135 0.527
p. 4). In o he wo ds, p(c) is ound by using he p esen a ion
p obabili y o he wo non- e ical a ge s as weigh s o he hi
a e and adding his o he p oduc o he 1-False ala m a e
(i.e., co ec ejec ion a e) and he p esen a ion p obabili y o he
e ical a ge [i.e., p(c) = (8/36)∗H + (16/36)∗H + (12/36)∗(1-F);
Macmillan and C eelman, 2004, p. 89].
Using his o mula, we obse ed a mean p(c) o 54 ±4.2%
(SE) in ials a ed as ully unawa e. He e he BF was
a he inconclusi e as o whe he p(c) was di e en om he
50% chance le el wi h only anecdo al e idence o he H0,
BF(10)= 0.409, 95% CI (44.9, 63.1) (Quin ana and Williams,
2018). In AL2 ials he mean p(c) on g oup le el was 77.1 ±2%
associa ed wi h a BF p o iding ex eme e idence ha p(c) was
uly abo e chance, BF(10)>100, 95% CI (72.6, 81.5) (Quin ana
and Williams, 2018). In ials a ed as almos ully awa e (AL3)
we obse ed a mean p(c) o 85.4 ±3.2%. He e he BF again
p o ided ex eme e idence o p(c) o be g ea e han 50%,
BF(10)>100, 95% CI (78.4, 92.4) (Quin ana and Williams,
2018). Violin plo s show he obse ed p(c) as a unc ion o
subjec i e awa eness in Figu e 3. Fo mo e anspa ency, we
addi ionally included accu acy da a ob ained in he expe imen al
ask, as well as he a e age a es o hi s (H), alse ala ms
(FA), co ec ejec ions (CR), and misses (M), and he mean
numbe o hi , alse ala m, miss, co ec ejec ion ials in he
Supplemen a y Ma e ial.
Taken oge he , using p(c) as a measu e o olun ee s’
pe cep ual sensi i i y did no change he conclusion ha
pa icipan s’ classi ica ion abili y was a chance in ials a ed
as subjec i ely ully unawa e, while hey showed conside able
classi ica ion sensi i i y in ials wi h esidual and almos ull
subjec i e awa eness. Impo an ly, he abo e measu es a e
ep esen a i e and exhaus i e o he c i ical a ge ea u e ha
is ele an o he ask (i.e., o ien a ion; Snodg ass e al.,
2004). Howe e , addi ional expe imen a ion could be pe o med
employing an e en mo e s ingen de ec ion h eshold in which
one’s sensi i i y o de ec he p esence o any g a ing is null.
RT Da a
We analyzed olun ee s’ RT da a o es whe he he la ency
o he manual esponses slowed down du ing (unconscious)
changes in he a ge o ien a ion compa ed o epea ing a ge
o ien a ions which would sugges a eweigh ing o a en ional
selec ion weigh s. Indi idual ou lie s (M±3SD) we e emo ed
be o e he LMM analysis. We conduc ed he same LMM analysis
o wo RT models. While in he i s a e age RT model he
swi ch condi ion comp ised all possible o ien a ion changes,
he second weigh ed RT model included only swi ch ials in
which he p io a ge o ien a ion was associa ed wi h a high
selec ion weigh (i.e., changes away om he mos equen
il ). Desc ip i e mean RTs and SEs o bo h models o swi ch
s. epea ials o each le el o awa eness a e summa ized in
Table 2.
To begin, we conduc ed he LMM analysis o he a e age
RT model in which he mean o he swi ch condi ion included
all possible swi ch ials. Visual inspec ion o esidual plo s did
no e eal any ob ious de ia ions om homoscedas ici y no
no mali y. Es ima ed RTs appea ed o be sensi i e o changes in
F on ie s in Human Neu oscience | www. on ie sin.o g 7Janua y 2021 | Volume 15 | A icle 610347
Güldene e al. Unconscious Re-weigh ing o Fea u e-Based A en ion
FIGURE 2 | Violin plo s o g oup dis ibu ions o sensi i i y A’ (le ) and he c i e ion loca ion ( esponse bias; igh ) o each le el o subjec i e awa eness [pe cep ual
awa eness scale (PAS) a ings AL1–AL3]. Le : he ag eemen o he objec i e and subjec i e measu e o isual awa eness is indica ed by mode a e e idence o he
absence o sensi i i y on ials a ed as subjec i ely unawa e; BF(10)<0.33. Black as e isks = BF(10)>100 indica ing ex eme e idence o A’ being uly >0.5.
Violin plo s use densi y cu es o depic dis ibu ions o nume ic da a. The wid h co esponds wi h he app oxima e equency o da a poin s in each egion. The
lowe and uppe limi s o each plo a e de e mined by he minimum and maximum alues.
FIGURE 3 | Violin plo s show he obse ed p opo ions o co ec esponses
[p(c)] as a unc ion o subjec i e awa eness (AL1–AL3). Black as e isks
indica e ha es ing p(c) on g oup le el agains a heo e ical chance le el o
0.5 (do ed line) esul ed in a Bayes ac o (BF) p o iding ex eme e idence o
p(c) being g ea e han 0.5 (BF(10)>100). Violin plo s use densi y cu es o
depic dis ibu ions o nume ic da a. The wid h co esponds wi h he
app oxima e equency o da a poin s in each egion. The lowe and uppe
limi s o each plo a e de e mined by he minimum and maximum alues.
he le el o isual awa eness indica ed by he signi ican ixed
e ec o awa eness, F(2,11.055) = 9.6740, p= 0.0037. In line wi h
ou p edic ions, he pos hoc es s showed ha RTs (a e aged
ac oss he condi ions swi ch and epea ) in AL1 ials we e
157.6 ±41.6 ms slowe compa ed o AL2 ials, (12.00) = 3.783,
p= 0.0078, 95% CI (41.8, 273.5), and 225.6 ±54 ms slowe
compa ed o AL3 ials, (11.87) = 4.177, p= 0.0039, 95% CI (75.2,
376.1). RTs in AL2 and AL3 ials did no di e signi ican ly,
p= 0.3678, 95% CI (−46.0, 182.0). Thus, RTs o he a e age RT
model was indeed sensi i e o changes in isual awa eness and
dec eased wi h inc easing s imulus isibili y.
The e was, howe e , no signi ican main e ec o swi ch,
F(1,35) = 3.1709, p= 0.0836, no a signi ican in e ac ion
F(2,35) = 0.1411, p= 0.8689 showing ha RTs appea ed o be
una ec ed by changing s imulus o ien a ions in his RT model.
Abou 20% o he o al a iance was explained by he model’s
ixed e ec s, R2
m= 0.1989, and 88% by he model’s ixed and
andom e ec s, R2
c= 0.8843.
Nex , we used he same LMM o analyze he weigh ed RT
model in which he swi ch condi ion comp ised only swi ch ials
whe e he p io o ien a ion was he hea ily weigh ed one. Again,
isual inspec ion o esidual plo s did no e eal any ob ious
de ia ions om homoscedas ici y no no mali y. The LMM
analysis showed, also in his model, ha es ima ed RTs inc eased
wi h dec easing isual awa eness, F(2,10.93) = 10.9895, p= 0.0024.
The pos hoc es s wi h Bon e oni co ec ion indica ed ha mean
RTs ac oss bo h swi ch and epea ials we e on a e age abou
190.5 ±45.6 ms signi ican ly slowe in AL1 ials compa ed
o AL2, (12) = 4.177, p= 0.0039, 95% CI (63.7, 317.3) and
on a e age 263.1 ±58.7 ms slowe compa ed o AL3 ials
(11.92) = 4.482, p= 0.0023, 95% CI (99.7, 426.4). Mean RTs in
AL2 and AL3 ials did no di e signi ican ly, p= 0.3007, 95%
CI (−40.8, 185.9).
Impo an ly, now also a swi ch o he a ge o ien a ion
a ec ed RTs: he analysis e ealed a signi ican ixed e ec
p edic o swi ch, F(1,35.00) = 6.0303, p= 0.019. He e he pos hoc
F on ie s in Human Neu oscience | www. on ie sin.o g 8Janua y 2021 | Volume 15 | A icle 610347
Güldene e al. Unconscious Re-weigh ing o Fea u e-Based A en ion
TABLE 2 | Mean (M) eac ion imes (RTs) and s anda d de ia ions (SD) o he swi ch and epea ials o each le el o subjec i e awa eness summa ized o (A) he
a e age RT model comp ising he mean o all swi ch ials and (B) he weigh ed RT model including only weigh ed swi ch ials.
Awa eness
Le el 1 Le el 2 Le el 3
Swi ch Repea Swi ch Repea Swi ch Repea
(A) A e age swi ch
M 1.110 1.068 0.945 0.917 0.889 0.869
SD 0.225 0.231 0.182 0.212 0.130 0.128
(B) Weigh ed swi ch
M 1.177 1.068 0.947 0.917 0.879 0.869
SD 0.242 0.231 0.194 0.212 0.129 0.128
es s sugges ed ha only in unawa e ials (AL1) we e RTs in
esponse o a no el o ien a ions on a e age 109.5 ±34.5 ms
signi ican ly slowe compa ed o ials in which he o ien a ion
was epea ed, (35) =−3.171, p= 0.0029, 95% CI (−179.7,
−39.4). In ials wi h highe le els o isual awa eness, swi ch
cos s we e no signi ican , AL2, p= 0.4033, 95% CI (−99.4,
40.9); AL3, p= 0.7789, 95% CI (−83.1, 62.8). Ye , he e
was no signi ican in e ac ion be ween he wo ixed-e ec
p edic o s awa eness and swi ch, F(2,35.00) = 2.2800, p= 0.1172.
Toge he , abou 26% o he o al a iance was explained by
he wo ixed e ec s awa eness and swi ch, R2
m= 0.2568, and
abou 85% was explained by all ixed and andom e ec s,
R2
c= 0.8532.
RTs o bo h swi ch and epea ials as a unc ion o isual
awa eness o he weigh ed and he exhaus i e RT model a e
plo ed in Figu es 4A,B. The LMM solu ions o he ixed and
andom e ec s o he wo RT models a e gi en in Tables 3A,B.
In sum, he LMM analysis sugges s ha no only we e
RTs sensi i e o dec easing isual awa eness bu also changes
in he s imulus o ien a ion. Howe e , RT cos s due o such
changes we e obse ed only in he weigh ed RT model which
included only hose swi ch ials in which he no el o ien a ion
changed away om he highly biased o ien a ion (highly equen
il ) os e ing he conclusion ha he p io isual selec ion
bias had boos ed beha io al swi ch cos s in esponse o a
change in he a ge o ien a ion. As signi ican swi ch cos s
we e obse ed in unawa e ials only, he impac o he p io
selec ion bias boos ing beha io al swi ch cos s du ing a en ional
e-selec ion appea ed o be mos p ominen in he ull absence o
isual awa eness.
Gi en ha unde unconscious condi ions we had ewe ials
included in he analysis, ou lie s could ha e a s onge e ec
on he esul s. Since 3 SD is no a igid cu o o ou lie s,
we, he e o e, epea ed he analysis wi h a 2 SD, and 2.5 SD
cu o bu he esul s did no change in e ms o signi ican
ixed e ec s. Mos ele an o ou esea ch ques ion, we ound
a signi ican swi ch e ec o he weigh ed RT da a model in
AL1 bu nei he in AL2 no in AL3 ials o all h ee cu o s.
We conduc ed u he con ol analyses ha a e epo ed in he
Supplemen a y Ma e ial in which we ma ched he numbe o
ials be ween AL1, AL2, and AL3 ials by andom sampling o
p o e ha he low amoun o AL1 ials could no accoun o he
obse ed swi ch e ec , and used he numbe s o ials ob ained
o he weigh ed swi ch ials a ed as ully unawa e (AL1) o
do a Bayesian-based p edic ion o show ha he swi ch cos s
in he weigh ed RT model we e no associa ed wi h indi idual
ial numbe s.
Finally, o ule ou he possibili y ha in e ial esponse
p iming ins ead o a en ional weigh ing could accoun o he
obse ed swi ch e ec , we epea ed he LMM analysis o he
weigh ed swi ch model a e emo ing all weigh ed swi ch ials
p eceded by ials in which he o ien a ion had been pe cei ed
consciously o some ex en (i.e., AL2, and AL3 ‘‘p e a ge ’’ ials).
This we did because in e ial esponse p iming is hough o
necessi a e awa eness o he s imulus in he p eceding ial
(e.g., Pe emen e al., 2013). Using he same LMM we ound
only a ma ginal swi ch e ec , F(1,42.301) = 4.0112, p= 0.0516, a
signi ican ixed e ec o isual awa eness, F(2,14.399) = 17.5605,
p<0.001, and a signi ican in e ac ion o he wo ixed e ec s
swi ch and awa eness, F(2,42.198) = 4.7763, p= 0.0134. The ixed
and andom e ec solu ions o his analysis a e gi en in Table 4.
Impo an ly, pai ed compa isons eplica ed ou p e ious inding
showing ha unawa e weigh ed swi ch ials we e signi ican ly
slowe compa ed o unawa e epea ial, (33.24) =−2.954,
p= 0.0057, 95% CI (−360.5, −66.4), while he e we e no
di e ences be ween swi ch and epea ials o AL2, p= 0.1157,
no o AL3, p= 0.2304.
DISCUSSION
When olun ee s engaged in ou disc imina ion ask o masked
g a ings, RTs we e sensi i e o o ien a ion changes. Howe e ,
signi ican swi ch cos s we e ob ained only i he selec ion weigh
o he p io o ien a ion was high (i.e., he highly equen
il ) and i he no el o ien a ion was unconsciously pe cei ed.
Impo an ly, ou c i e ia o lack o awa eness we e based on
he combina ion o subjec i e and objec i e measu es, i.e., no
expe ience epo s and no abili y o disc imina e he ele an
a ge ea u es in a o ced-choice es . To he bes o ou
knowledge, his, he e o e, is he i s s udy in es iga ing he
e ec s o unawa e a ge s on ea u e-based a en ion weigh ing
by using a combina ion o objec i e sensi i i y measu es and
subjec i e measu es o isual (un-)awa eness collec ed du ing
he expe imen al ask. This is a e y impo an ad an age o
wo easons: i s , o accoun o luc ua ions o he pe cep ual
h eshold be o e, du ing, and a e he ac ual expe imen al ask
i is ex emely impo an o use an ‘‘online’’ measu e o isual
awa eness du ing he ask pe o mance. This way, one ensu es
F on ie s in Human Neu oscience | www. on ie sin.o g 9Janua y 2021 | Volume 15 | A icle 610347
