You are on page 1of 8

Los art?culos se proporcionan ?nicamente para uso personal y educativo. Est? prohibida la reproducci?n, distribuci?n y uso comercial.

Psychiatry Research 255 (2017) 347354

Contents lists available at ScienceDirect

Psychiatry Research
journal homepage:

Facial emotion recognition and borderline personality pathology MARK

a,d, b a c b
Kevin B. Meehan , Chiara De Panlis , Nicole M. Cain , Camilla Antonucci , Antonio Soliani ,
John F. Clarkind, Fabio Sambataroe
Department of Psychology, Long Island University, 1 University Plaza, Brooklyn, NY 11201, USA
Unit of Psychiatry, Department of Neuroscience, University of Parma, c/o Ospedale Maggiore, pad. 21 Braga, viale Gramsci 14, 43126 Parma, Italy
Service of Hospital-based Psychiatry, Department of Mental Health, Parma Local Health Agency, c/o Ospedale Maggiore, pad. 21 Braga, viale Gramsci 14, 43126 Parma,
Department of Psychiatry, Weill Cornell Medical College, 21 Bloomingdale Road, White Plains, NY 10605, USA
Department of Experimental and Clinical Medical Sciences, University of Udine, Udine, Italy


Keywords: The impact of borderline personality pathology on facial emotion recognition has been in dispute; with impaired,
Facial emotion recognition comparable, and enhanced accuracy found in high borderline personality groups. Discrepancies are likely driven
Borderline personality by variations in facial emotion recognition tasks across studies (stimuli type/intensity) and heterogeneity in
Eortful control borderline personality pathology. This study evaluates facial emotion recognition for neutral and negative
Social cognition
emotions (fear/sadness/disgust/anger) presented at varying intensities. Eortful control was evaluated as a
moderator of facial emotion recognition in borderline personality. Non-clinical multicultural undergraduates (n
Negative aect
= 132) completed a morphed facial emotion recognition task of neutral and negative emotional expressions
across dierent intensities (100% Neutral; 25%/50%/75% Emotion) and self-reported borderline personality
features and eortful control. Greater borderline personality features related to decreased accuracy in detecting
neutral faces, but increased accuracy in detecting negative emotion faces, particularly at low-intensity thresh-
olds. This pattern was moderated by eortful control; for individuals with low but not high eortful control,
greater borderline personality features related to misattributions of emotion to neutral expressions, and en-
hanced detection of low-intensity emotional expressions. Individuals with high borderline personality features
may therefore exhibit a bias toward detecting negative emotions that are not or barely present; however, good
self-regulatory skills may protect against this potential social-cognitive vulnerability.

1. Introduction been associated with, respectively, an under-attention to emotional

cues in faces; an empathy paradox in which those with BPD are hy-
Borderline Personality Disorder (BPD) is characterized by sig- persensitive to others emotions; and a negativity bias toward inter-
nicant interpersonal dysregulation that may be fostered by biased preting neutral faces as emotional. We identied two areas in prior
perceptions of social cues (Lieb et al., 2004). Decoding facial expres- research that can be addressed to clarify these seeming contradictions.
sions is an important aspect of social communication that inuences
how individuals respond to interpersonal contexts (Wagner and 1.1. Detection versus labeling of facial emotions
Linehan, 1999). However, studies evaluating facial emotion recognition
(FER) in BPD have collectively yielded inconsistent ndings. Studies Facial emotion recognition is a multifaceted process, involving both
have variously reported either impaired (Levine et al., 1997; Bland the detection of whether an emotion is present and the labeling/iden-
et al., 2004; Guitart-Masip et al., 2009; Unoka et al., 2011) or increased tication of that emotion (Daros et al., 2014). It has been proposed that
facial emotion recognition accuracy (Lynch et al., 2006; Minzenberg the former assessment i.e., discerning an emotion's presence in facial
et al., 2006; Domes et al., 2008; Dyck et al., 2009; Merkl et al., 2010; expressions: is that an emotion? can be intact or even more accurate
Schulze et al., 2013; Veague and Hooley, 2014) in BPD, while other in BPD, thus explaining the clinical observation of an empathy
studies indicated a tendency of those with BPD to misattribute negative paradox in BPD (i.e., those with BPD may be attending to and over-
emotions to neutral faces (Wagner and Linehan, 1999; Dyck et al., interpreting eeting emotions best left ignored). Conversely, the latter
2009; Daros et al., 2014). Therefore, BPD interpersonal dysfunction has process i.e., distinguishing among negative emotion labels: what

Correspondence to: Long Island University - Brooklyn Campus, 1 University Plaza, Brooklyn, NY 11201, United States.
E-mail address: (K.B. Meehan).

Available online 30 May 2017

Received 21 November 2016; Received in revised form 23 May 2017; Accepted 28 May 2017

0165-1781/ 2017 Elsevier B.V. All rights reserved.

Los art?culos se proporcionan ?nicamente para uso personal y educativo. Est? prohibida la reproducci?n, distribuci?n y uso comercial.

K.B. Meehan et al. Psychiatry Research 255 (2017) 347354

emotion is that? might be disrupted, thus accounting for BPD mis- Robin et al., 2012; Lowyck et al., 2016). For instance, the nding of
reading of social stimuli and emotional instability in interpersonal lower facial emotion recognition accuracy for full intensity negative
contexts (Daros et al., 2013, 2014; Schulze et al., 2013; Mitchell et al., emotions in BPD may be explained by bottom-up impairments in cog-
2014). However, some caveats apply to previous research in this area. nitive control: the hyperarousal evoked by highly salient emotional
First, research on pure facial emotion detection in BPD is still stimuli deplete cognitive resources and interfere with facial emotion
limited. In fact, most studies used dynamically morphing tasks asking to identication (Daros et al., 2013). Consistently, the studies (Gardner
identify when an emotion is recognizable as it unfolds from neutral to et al., 2010; Preti et al., 2016) that explicitly evaluated the interaction
full emotional intensity, and then label that emotion (Lynch et al., between eortful control, BPD features, and accuracy in labeling full-
2006; Domes et al., 2008; Robin et al., 2012;), or still morphing tasks intensity or dynamically morphed facial emotions found that, in those
prompting for emotion labeling (i.e., Veague and Hooley, 2014). In this with high BPD traits, good eortful control favors while low eortful
way, the two processes of facial emotion detection and identication control reduces facial emotion recognition accuracy. However, to date
are conated, leaving unclear whether impairments in accuracy are eortful control has not been directly investigated as a putative mod-
driven by altered detection or altered labeling of emotional cues. In erator in the association between BPD features and response accuracy
order to address this confound, Schulze et al. (2013) asked participants in both facial emotion detection and identication for low-intensity
to indicate the presence of an emotional stimulus within a stream of emotions.
neutral faces without the need to label the emotional state, and found
increased detection of emotional expressions in BPD. However, that 1.3. The present study
study employed only two FEs (100% intensity anger and happiness),
leaving it unclear whether BPD is associated with over-attending to The present study examined facial emotion recognition accuracy
subtle (i.e., less than 100% intensity) expressions of facial emotions, as using two separate tasks for evaluating emotion detection and labeling
the empathy paradox seems to imply. Therefore, the present study in a nonclinical sample with varying BPD traits. In the present tasks, all
evaluated facial emotion detection separately from labeling using two faces depict neutral and negative emotions. First, the detection task
dierent tasks to ensure that detection is not biased by expectancies of requires a rapid forced choice between neutral and emotion in faces
subsequently being asked to label the emotion that has just been de- presented across of range of emotional intensities (100% neutral and
tected. Further, the present detection task required discrimination of 25%/50%/75% emotion intensity) and valences (anger, fear, sadness,
neutral vs. emotion when viewing still images presented at varying disgust). Second, the labeling task allows for a more deliberative forced
emotional intensities (100% neutral and 25%/50%/75% emotion in- choice among valences (neutral, anger, fear, sadness, disgust), with the
tensities) to clarify whether BPD is related to a decreased threshold emotions only presented at a very low (25%) intensity. Self-reported
for detecting facial emotions. eortful control and borderline personality features then predicted
Second, studies evaluating labeling of full-intensity still facial detection and labeling accuracy.
emotions in BPD reported decreased accuracy for negative emotions, In keeping with the negativity bias and empathy paradox hy-
especially for rejection-related emotions such as anger and disgust potheses of facial emotion recognition in BPD, for both tasks we hy-
(Daros et al., 2013; Mitchell et al., 2014). Conversely, studies that used pothesized a bias towards over-attributing negative emotions to neutral
dynamic facial morphs reported that those with BPD or high BPD fea- or ambiguous facial expressions in those with high borderline person-
tures correctly identied facial emotions either at an earlier (Lynch ality features, particularly in those who also reported low eortful
et al., 2006), at a later (Robin et al., 2012), or at a comparable stage of a control. In other words, we expected those with borderline personality
dynamic morph than controls (Domes et al., 2008). Importantly, a dy- features to show a reduced detection and labeling accuracy for neutral
namic morph may actually limit the ability to inform about facial faces, suggesting that they interpret neutral expressions as emotional,
emotion recognition at very low intensities, because in practice re- and an enhanced detection and labeling accuracy for emotional faces
sponses tend to occur as the emotion has moved to the higher intensity presented at very low (25%) intensity, suggesting that they are hy-
range (broadly in the 4575% diusion range across studies) when the persensitive to subtle expressions of negative emotions in others (pos-
emotion has become clear enough to be accurately labeled. Only one sibly with stronger eects for emotions that connote rejection, such as
study (Daros et al., 2014) evaluated the lower intensity range with still anger and disgust). We further expected that both of these patterns (i.e.,
images at low diusion (as well as medium/high), and found no dif- the tendency to misinterpret neutral stimuli as emotional coupled with
ferences in terms of BPD status, but only sadness and happiness were an increased ability to detect and label ambiguous, low-intensity ne-
evaluated, and the percent intensity of these low diusion images were gative facial emotions) would be buered by high eortful control.
not specied. Thus, the present study evaluated labeling accuracy for
still images of neutral and low-intensity (25%) negative emotional ex- 2. Methods
pressions; with no full (100%) intensity emotions presented.
2.1. Participants and procedures
1.2. The role of self-regulatory skills
In total, 186 undergraduates from a multicultural U.S. university
Discrepancies in past facial emotion recognition ndings may not be participated in a study approved by the university's institutional review
explained by BPD alone, but through its interaction with other per- board. Prior to the study, the Benton Facial Recognition Test short-form
sonality variables that can inuence social-cognitive functioning. (BFRT) was administered to exclude facial emotion recognition decits
Eortful Control (EC) is the self-regulatory aspect of temperament that due to impaired recognition of facial identity (prosopagnosia); scores
allows individuals to strategically regulate contingent emotions, im- 39 were classied as normal (Benton et al., 1994). One participant was
pulses and thoughts for the sake of valued goals, thereby promoting excluded for BFRT performance; the sample mean was 46.09 (SD =
social adjustment (Rothbart et al., 2011; De Panlis et al., 2013). Ef- 3.70). Participants then completed the facial emotion recognition tasks
fortful control buers the link between relational vulnerabilities and and self-report measures, and were compensated with course credit.
BPD features (Ayduk et al., 2008; De Panlis et al., 2016) and dier- Fifty-four participants (32 females) were removed prior to analyses
entiates between high and low functioning subgroups of BPD patients because they failed multiple attention checks embedded in the protocol
(Hoermann et al., 2005). Notably, eortful control reects the e- (i.e., questionnaire items that ask for a specic response based on a
ciency of executive attention (Posner and Rothbart, 2009), and execu- close read of the item; see Goodman et al., 2013). The analyses pre-
tive/attentional control has been proposed as an inuential determi- sented here are based on the remaining (n = 132) participants. How-
nant of facial emotion recognition in BPD (e.g., Domes et al., 2009; ever, participants could skip items they did not feel comfortable

Los art?culos se proporcionan ?nicamente para uso personal y educativo. Est? prohibida la reproducci?n, distribuci?n y uso comercial.

K.B. Meehan et al. Psychiatry Research 255 (2017) 347354

responding to, resulting in missing data on some measures (see below). with an inter-trial interval of 1 s. Stimuli were pseudo-randomized by
Participants (76.5% female) ranged in age from 18 to 53 years emotion valence (neutral, anger, disgust, fear, sadness) and emotion
(mean age 21.39 6.38), with 80% between 18 and 22 years of age. intensity (25/50/75%) to reduce habituation and expectancy eects. A
The ethnic composition of the sample was 33.3% African-American, total of 450 trials were presented, lasting 20 min. Accuracy and
18.5% Asian, 17.8% Latino, 17.8% Caucasian, 7.4% Middle Eastern, reaction time were recorded.
and 4.4% Other. Labeling task. In order to clarify accuracy of recognition of
2.2. Measures low-intensity emotions, the detection task was followed by a labeling
task evaluating neutral and negative facial emotions (anger, disgust,
2.2.1. Facial emotion recognition tasks fear, sadness) at a low level (25%) of intensity. Subjects determined by Stimuli preparation. Both the facial emotion detection and multiple choice which expression among ve valences (i.e. neutral,
labeling tasks included neutral and negative (sadness, fear, anger, sadness, fear, anger, disgust) is portrayed by the face (Fig. 1). The total
disgust) facial stimuli taken from Ekman's Japanese and Caucasian number of trials was 70 (30 neutral stimuli, 40 emotion stimuli, i.e., 10
Facial Expressions of Emotion (JACFEE) and Neutral Faces (JACNeuF) stimuli for each emotion). Stimuli were pseudorandomized by emotion
series (Ekman and Friesen, 1976; Matsumoto and Ekman, 1988) and type. There were no response time limit: the next stimulus would
NimStim (Tottenham et al., 2009). Face morphs of both sexes were appear only when the subject had made a choice on the preceding one.
created with Abrosoft Fantamorph 5.4 by blending negative and neutral Again, the performance measures were accuracy and reaction time.
expressions of the same identity in 25% increments (LaBar et al., 2003)
to obtain four levels of emotional intensity (100% neutral and 25%/
2.2.2. Questionnaires
50%/75% emotion intensity). SCID-II screener for personality disorders: BPD scale. The SCID-II
questionnaire (First et al., 1997) was constructed as a screener that Detection task. This task evaluated detection of neutral and precedes the full SCID-II semi-structured interview, with Yes/No
negative facial emotions (anger, disgust, fear, sadness) across dierent statements referring to the presence/absence of DSM-IV personality
levels of intensity (25%/50%/75% emotion intensity). Subjects were disorder symptoms. Using the SCID-II Interview as gold standard, the
required to indicate, as quickly and accurately as possible, whether the SCID-II questionnaire has demonstrated good psychometrics as a
face displayed an emotion or not. Two labels, neutral and emotion, measure of personality disorder pathology in its own right, with good
were presented under a morphed face, prompting a forced choice by criterion validity (Germans et al., 2010) construct validity (Piedmont
pressing the left or right mouse button (Fig. 1). Each trial lasted for 2 s et al., 2003) and clinical utility in assessing BPD (Chanen et al., 2008).
In this study we administered only the BPD scale, which consists of 15
questions covering the nine DSM-IV BPD criteria. The total number of
BPD items endorsed is the index of severity of borderline personality
features, which was found to have high internal validity ( = .85). Eortful control (EC). Eortful control was assessed with the
19-item EC scale of the Adult Temperament Questionnaire (ATQ; Evans
and Rothbart, 2007). Items on a 7-point Likert-scale (1 = extremely
untrue of you, to 7 = extremely true of you) refer to inhibitory control,
activation control, and eortful attention. The internal consistency of
the EC scale in the present sample was = .68, which is only adequate
but closely compares to other studies using a larger sample of a similar
cohort ( = .71; Cain et al., 2013) and to the validation study among
undergraduates ( = .78).

2.3. Statistical analysis

Reaction time (RT) was evaluated as a potential covariate for la-

beling, but not for detection given that not responding within 2 s is a
ceiling past which the response is deemed inaccurate, there was an
expected negative relationship between accuracy and RT (rneutral =
.27; remotion = .31) that indicates no speed-accuracy trade-o.
Potential covariates were evaluated with Pearson correlations (age) and
t-tests (sex). The relationship between borderline personality features
and both detection and labeling accuracy were evaluated with Pearson
correlations and linear regression models (SPSS.20), with the latter
meeting the assumption of independent errors (Durbin-Watson, all d's
= 1.68 2.14). Further, the interaction of eortful control in the re-
lationship between borderline personality features and accuracy was
probed with moderation analysis using PROCESS for SPSS (Hayes,
2013). Because emotion detection accuracy varied by intensity per-
centage, multilevel linear modeling (MLM) was utilized (in STATA.14)
with a restricted maximum-likelihood random-eects model. For these
models emotion detection accuracy scores (Level 1) that varied across
Fig. 1. Screenshots illustrating the facial emotion recognition detection (A) and labeling intensity percentage were nested within individuals, and eortful
(B) tasks. Both tasks were presented on a Windows computer, using E-Prime 2.0 software, control and borderline personality features were evaluated as (Level 2)
and were preceded by detailed instructions. moderators.

Los art?culos se proporcionan ?nicamente para uso personal y educativo. Est? prohibida la reproducci?n, distribuci?n y uso comercial.

K.B. Meehan et al. Psychiatry Research 255 (2017) 347354

Table 1
Descriptive statistics and pearson correlations between accuracy, borderline personality
features, and eortful control.


Detection: % Accurate M (SD) r r

Neutral 78.18 (16.80) .21* .06
Emotion 73.14 (10.56) .24* .09
25% Emotion 46.37 (18.20) .24* .09
50% Emotion 82.61 (10.31) .21* .09
75% Emotion 90.43 (7.09) .14 .05
Anger 71.60 (11.91) .19* .08
Disgust 85.43 (7.73) .24* .11
Fear 68.00 (13.67) .30* .09
Sadness 67.52 (13.28) .15 .06
Labeling: % Accurate
Neutral 89.22 (9.93) .28* +
Anger 38.64 (24.11) .21* +
+ +
Disgust 34.55 (17.62) .14 .31*
Fear 33.64 (19.51) .21* +
+ +
Sadness 31.44 (21.05) .04 .05
Reaction Time (ms) 2268.38 (885.26) .17 .02
M (SD) M (SD)
5.13 (3.74) 4.45 (.72)

Note. Responses were obtained from 132 participants. On the EC scale, 11 subjects had
partially missing data, mean-replacement was used on those with less than 10% missing
Fig. 2. Detection accuracy for neutral faces predicted by borderline personality features
data, resulting in n = 129. On the SCID-II borderline personality scale, 19 subjects had
in interaction with eortful control. Note. The moderator (eortful control) is continuous,
partially missing data; because symptoms are discrete mean replacement is not appro-
with its conditional eect depicted at one standard deviation above (high = 5.17), below
priate, resulting in n = 113. BP = Borderline Personality features on the SCID-II ques-
(low = 3.76), and at the mean (medium = 4.46). The moderator value dening the
tionnaire; EC = Eortful Control on the Adult Temperament Questionnaire; +partial
Johnson-Neyman signicance region is below 4.33, which encompasses the region of low
correlation controlling for Reaction Time.
to medium eortful control.
* p < .05.

neutral faces was found with high borderline personality features for
3. Results
those with low but not high eortful control ( = 1.22, p = .03),
though the eect was small (F[3,108] = 3.31, p = .02, R2 = .08). The
Responses were obtained from 132 participants. On the EC scale, 11
Johnson-Neyman technique indicated that for individuals scoring
subjects had partially missing data, mean-replacement was used for
higher than 4.33 on the EC scale (i.e., just below the mean; 56% of the
those with less than 10% missing data, resulting in n = 129. On the
sample) borderline personality features had no signicant eect on
SCID-II borderline personality scale, 19 subjects had partially missing
neutral detection accuracy (Fig. 2). Therefore, only those with high
data; as symptoms are discrete mean replacement is not appropriate,
borderline personality features and low eortful control displayed re-
resulting in n = 113.
duced accuracy for neutral faces.

3.1. Detection of facial expressions 3.1.2. Emotion detection accuracy

Eortful control was also evaluated as a moderator of the re-
Detection accuracy for the two response options (neutral vs. lationship between borderline personality features and emotion detec-
emotion) was analyzed separately; the emotion response option tion accuracy (across levels of intensity). A signicant interaction was
included faces that varied by valence and intensity, and thus accuracy found, such that greater accuracy for emotion faces was found with
was probed across these variations. The relationship of demographic high borderline personality features for those with low but not high
variables to detection accuracy was evaluated; sex and age were not eortful control ( = 1.77, p = .001), with a medium eect (F
related to emotion (tsex = 1.34, p = .18; rage = .02, p = .83) or neutral [3,108] = 6.03, p = .001, R2 = .14). The Johnson-Neyman technique
(tsex = 1.09, p = .28; rage = .01, p = .88) accuracy. Eortful control indicated that for individuals scoring higher than 4.43 on the eortful
was not related to neutral or total emotion detection accuracy across control scale (i.e., at the mean; 51% of the sample) borderline per-
any valence or intensity (see Table 1). However, greater borderline sonality features had no signicant eect on emotion detection accu-
personality features was signicantly correlated with lower neutral (r racy (Fig. 3). Therefore, only those with high borderline personality
= .21, p = .03) and higher emotion (r = .24, p = .01) detection features and low eortful control displayed enhanced accuracy for
accuracy, and this higher accuracy was observed when the emotion emotion faces.
depicted was anger (r = .19, p = .046), fear (r = .30, p = .001), and MLM was used to evaluate the moderating role of emotion intensity,
disgust (r = .24, p = .01), but not sadness (r = .15, p = .12). The given that accuracy for emotion responses was evaluated across three
positive relationship between borderline personality features and de- degrees of intensity (Level 1) as a function of interacting baseline levels
tection accuracy was observed when the emotion was presented at low of borderline personality and eortful control (Level 2). Consistent with
(25%; r = .24, p = .01) and medium (50%; r = .21, p = .03) but not above, in Model 1 there was a signicant and positive main eect of
high (75%; r = .14, p = .16) intensity. These relationships were Intensity Percentage as well as borderline personality features on
therefore probed with models that included eortful control as a greater emotion detection accuracy, and in Model 2 there was a sig-
moderator. nicant interaction between eortful control and borderline personality
features (Table 2). In Model 3, there was an interaction between bor-
3.1.1. Neutral detection accuracy derline personality features and Intensity Percentage, such that the
Eortful control was evaluated as a moderator of the relationship impact of greater borderline personality features on greater emotion
between borderline personality features and neutral detection accuracy; detection accuracy is observed at low (25%) but not high (75%) in-
a signicant interaction was found, such that reduced accuracy for tensity. This interaction is further moderated by eortful control: the

Los art?culos se proporcionan ?nicamente para uso personal y educativo. Est? prohibida la reproducci?n, distribuci?n y uso comercial.

K.B. Meehan et al. Psychiatry Research 255 (2017) 347354

impact of greater borderline personality features on greater low in-

tensity detection accuracy is observed in those with low but not high
eortful control (all p < .01; Fig. 4). Therefore, only those with high
borderline personality features and low eortful control displayed en-
hanced accuracy for subtle presentations of emotion in faces.

3.2. Labeling of facial expressions

In order to clarify accuracy of recognition of low-intensity emotions,

labeling accuracy for the ve response options (neutral, anger, fear,
disgust, sadness; with emotions all at 25% intensity) were analyzed
separately. RT was evaluated as a potential covariate and found to be
negatively related to neutral labeling accuracy (r = .47, p = .001)
and positively related to anger (r = .35, p = .001), fear (r = .25, p =
.004), sadness (r = .37, p = .001), but not disgust (r = .15, p = .10).
This suggests a speed-accuracy trade-o for labeling emotions, with
accurate responses only at the expense of a longer RT, whereas for
neutral stimuli speed and accuracy are working in concert.
Demographic variables were also evaluated; age was not related to la-
beling accuracy for neutral (r = .04), anger (r = .01), disgust (r =
.14), fear (r = .07), or sadness (r = .05, all ns). Female sex was
related to labeling accuracy for neutral (t = 2.18, p = .03), anger (t
= 2.77, p = .01), and disgust (t = 2.38, p = .02), but not fear (t = .44,
Fig. 3. Detection accuracy for emotion faces predicted by borderline personality features
p = .66) and sadness (t = .53, p = .60).
in interaction with eortful control. Note. The moderator (eortful control) is continuous,
with its conditional eect depicted at one standard deviation above (high = 5.17), below Greater borderline personality features were signicantly correlated
(low = 3.76), and at the mean (medium = 4.46). The moderator value dening the with lower neutral (r = .28, p = .004) and higher emotion labeling
Johnson-Neyman signicance region is below 4.43, which encompasses the region of low accuracy for anger (r = .21, p = .02) and fear (r = .21, p = .03), but
to medium eortful control. not disgust (r = .14) and sadness (r = .04, both ns) (Table 1), con-
trolling for RT. Greater eortful control was related to less accurate
Table 2 emotion labeling of disgust (r = .30, p = .001) and fear (r = .19, p
Multilevel model of emotion detection accuracy by intensity. = .045), but was not related to sadness (r = .05) and anger (r =
.14), nor was it related to neutral labeling (r = .14, all ns). These
Model 1 Model 2 Model 3
relationships were probed with linear regression models that included
** **
Intensity% 22.63 (.80) 22.60 (.80) 39.09 (9.18)** eortful control as a moderator, controlling for RT. However, no sig-
BP .65 (.25)* 5.76 (1.59)** 12.92 (3.19)** nicant interaction terms were found in models evaluating the inter-
EC 5.55 (2.21)* 11.81 (4.46)*
action of borderline personality features and eortful control in pre-
BP EC 1.18 (.36)** 2.64 (.73)**
% BP 3.58 (1.38)* dicting the labeling accuracy of neutral ( = .18, p = .73), anger (
% EC 3.13 (1.93) = .62, p = .27), disgust ( = 1.05, p = .06), fear ( = .05, p = .93),
% BP EC .73 (.32)* or sadness ( = .42, p = .44).
Intercept 51.56 (14.11) 43.62 (12.49) 45.08 (13.23) Exploratory analyses evaluated whether there was a pattern to the
Residual 144.11 (13.59) 143.97 (13.63) 139.56 (13.30)
2LL 1362.14 1343.16 1337.93
inaccurate labeling related to greater borderline personality features.
For errors on emotion faces, those higher on borderline personality
Note. BP = Borderline Personality features on the SCID-II questionnaire; EC = Eortful were less likely to mistake faces depicting anger (r = .27, p = .01),
Control on the Adult Temperament Questionnaire. fear (r = .34, p = .00), disgust (r = .21, p = .04), and sadness (r =
* p < .01. .28, p = .00) as neutral, controlling for RT. Put dierently, those with
** p < .001.
higher borderline personality features were more likely to inaccurately
label an emotion as another valenced emotion, rather than a neutral
face, though there was no specic pattern with regard to what valenced
emotion was instead attributed. For errors on neutral faces, those with
higher borderline personality features were more likely to inaccurately
attribute sadness to neutral faces (r = .22, p = .03), but not other
valences (ranger = .02; rdisgust = .15; rfear = .02; all ns).

4. Discussion

4.1. Detection of facial expressions

The present study rst assessed the degree to which an emotion is

detected to be present at all even if too eeting to be categorized and
labeled by requiring a quick, dichotomous distinction of emotion
versus neutral, with emotions randomized across valences and in-
tensities. Those with high borderline personality features were less
accurate at detecting neutral faces, and more accurate at detecting
subtle, low-intensity negative emotions in faces.
Fig. 4. Detection accuracy for emotion faces varies by borderline personality features in
interaction with eortful control and intensity %.
With respect to the reduced detection of neutral faces, these results
extend previous ndings of inaccurate attributions of an emotional

Los art?culos se proporcionan ?nicamente para uso personal y educativo. Est? prohibida la reproducci?n, distribuci?n y uso comercial.

K.B. Meehan et al. Psychiatry Research 255 (2017) 347354

valence to neutral expressions in those with BPD (Daros et al., 2013) full-intensity emotions.
and further clarify that this bias towards seeing emotions in neutral
faces is apparent even in the detection phase of facial emotion re- 4.3. The moderating role of eortful control
cognition, irrespective from the potential confounds driven by the need
to choose among diverse emotion labels. This study was the rst to evaluate eortful control as a potential
With respect to the increased detection of emotional stimuli, the moderator of both low-intensity facial emotion detection and labeling.
present results extend the ndings of Lynch et al. (2006), who found With regard to the detection phase of facial emotion recognition, good
that those with BPD correctly identied emotions at an earlier stage of a self-regulatory skills (i.e., high eortful control) buered against the
dynamic morph than controls across a range of positive and negative automatic response tendency of those with high borderline personality
valences, as well as those of Schulze et al. (2013), who reported en- features to see emotions in faces with no (neutral) or subtle (25%)
hanced detection of 100% emotional facial expressions presented signs of emotions. Notably, emotion detection is regarded as an auto-
among rapidly changing neutral facial stimuli. However, those studies matic and reexive process (Adolphs, 2002; Phillips et al., 2008). In
left unclear what kinds of preliminary impressions may have been those with high borderline personality features this bias toward a self-
present at low-intensity emotion diusions, while the present facial referential and negative processing of social cues (i.e., a misattribution
emotion detection specically claried that there is a bias at the lower of emotion to neutral faces coupled with enhanced detection of very
level intensity of emotion detection. Notably, higher borderline per- subtle negative emotions in faces) is buered by high eortful control.
sonality features were related to more accurate emotion detection when This conrms the role of eortful control in protecting against BPD-
the valence depicted indicated potential interpersonal rejection (i.e., related social-cognitive vulnerabilities (Ayduk et al., 2008; De Panlis
anger and disgust) or threat/alarm around or about oneself (i.e., fear), et al., 2016) and in promoting interpersonal adjustment: in the service
but not sadness. In fact, sadness potentially conveys less salient in- of social bonding and functioning, it may actually be adaptive to not
formation about the high borderline personality respondent it is more over-attend to eeting negative expressions in other's faces.
likely to elicit concern for the other person than signal danger for the Eortful control did not moderate the association between high
self. borderline personality traits and emotion identication through verbal
Taken together, the results on the detection phase of facial emotion labeling, which is regarded as a more voluntary, reective process
recognition indicate that those high in borderline personality features than emotion detection (Adolphs, 2002; Phillips et al., 2008). Thus, the
exhibit a lower detection threshold for seeing emotion in faces, either moderating role of eortful control may be more evident at the more
inaccurately (e.g., misperceiving a neutral face as emotional) or accu- reexive detection phase, through providing adequate top-down
rately (i.e., detecting low-intensity, ambiguous facial emotions as va- control of bottom-up automatic response tendencies, rather than during
lenced, especially if these signal threat or rejection). the more reective, deliberative labeling phase of neutral or low-in-
tensity facial emotions. This is consistent with the conceptualizations of
4.2. Labeling of facial expressions the centrality of eortful control in the modulation of reactive ten-
dencies (Rothbart and Bates, 1998). However, greater eortful control
The labeling task subsequently claried this detection bias by may buer the relationship between greater borderline personality
evaluating the ability to correctly identify subtle emotional expressions features and less accurate labeling of angry expressions presented at
across ve valenced choices (anger, fear, sadness, disgust, and neutral). higher intensities (Gardner et al., 2010; Preti et al., 2016), which are
While this task is procedurally closer to past studies evaluating facial likely to deplete cognitive resources to a greater extent than low-in-
emotion recognition with still images, it was the rst to clarify accuracy tensity ones. It may be that good eortful control may protect those
of recognition of low-intensity (25%) emotional stimuli, which is im- with high borderline personality features from the arousal evoked by
portant given that biased detection was most evident at this level high-intensity emotional faces, thereby favoring their accurate identi-
(whereas the sample was largely accurate at detecting 75% emotions). cation. Conversely, the low-intensity emotions displayed in the cur-
Consistent with expectations, greater borderline personality features rent study may attract BPD individuals attention toward salient social
predicted reduced labeling accuracy of neutral faces but greater la- stimuli in the environment without increasing one's arousal, thus fa-
beling accuracy of low-intensity angry and fearful faces. voring their enhanced deliberative processing irrespective of top-down
The nding of a decreased ability to correctly label neutral faces cognitive control (Daros et al., 2013). The role of arousal evoked by
extends the results of the detection phase of facial emotion recognition: facial expressions would need to be borne out in future research.
the tendency towards inaccurate attributions of negative emotions to
neutral faces exhibited by those with high borderline personality fea- 4.4. Implications for BPD interpersonal dysfunction
tures does not diminish when provided with greater time for reection,
nor when provided with explicit labels for emotions. Notably, the Overall, these results do not indicate a general facial emotion re-
sample overall was quite accurate in labeling neutral faces, with much cognition decit in BPD, but are consistent with recent research sug-
higher rates of errors occurring for low-intensity emotional stimuli; gesting that while some processes may be impaired (i.e., the accurate
however, high borderline personality features were specically asso- recognition of neutral faces), others may be intact or even amplied
ciated with a negativity bias toward interpreting neutral expressions (i.e., the accurate recognition of low-intensity negative emotions,
as emotional, consistent with previous studies (Daros et al., 2013). especially those signaling rejection or danger) (Hidalgo et al., 2016).
With respect to the labeling of low-intensity emotions, those with Thus, ndings are consistent with both a negativity bias, by which
high borderline personality features were signicantly less likely to those with high borderline personality features misinterpret neutral
mistake an emotion for a neutral face. Put dierently, those with low social stimuli as negative, and an empathy paradox in which they
borderline personality features were signicantly more likely to fail to show an enhanced attention toward salient social stimuli (Dinsdale and
notice a subtle emotion and mislabel it neutral. While this response Crespi, 2013). These tendencies towards greater detection of emotions
pattern is consistent with past studies nding heightened sensitivity to that are either not present (neutral) or barely present (25% emotion)
negative facial emotions in those with BPD (Domes et al., 2008), pre- may each contribute to detriments in social functioning. Inaccurately
vious studies also reported a decreased accuracy in identifying full-in- detecting eeting negative emotions in neutral faces may be the basis of
tensity (100%) negative emotions (Daros et al., 2013; Mitchell et al., distorted attributions that lead to interpersonal ruptures. Accurate de-
2014). This suggests an interesting facial emotion recognition pattern in tection of subtle negative cues of anger and fear may also be mala-
BPD, such that an enhanced identication of subtle, ambiguous signs of daptive; a good deal of coordinated interpersonal exchange involves not
emotions is coupled with increasing diculties in correctly identifying over-noticing every aect cue that signals rejection or danger in the

Los art?culos se proporcionan ?nicamente para uso personal y educativo. Est? prohibida la reproducci?n, distribuci?n y uso comercial.

K.B. Meehan et al. Psychiatry Research 255 (2017) 347354

service of keeping up positive social relatedness. It may even be Because our primary interest was in evaluating a negativity bias in
adaptive to at times mistake subtle negative expressions for neutral borderline personality pathology, we only evaluated neutral and ne-
ones in the context of a warm, mutual social exchange a subtle ash of gative emotion faces. However, BPD has been associated with specic
anger or disgust might not have much heuristic value but in this study diculties in categorizing positive, 60% intensity happy facial emo-
those with high borderline personality did not tend to make mistakes in tions (Fenske et al., 2015); therefore future studies should also evaluate
that direction. Thus, the fear of rejection may lead those with high facial emotion detection and labeling for low-intensity positive facial
borderline personality features to over-attend signals of danger or stimuli, as well as whether good eortful control could buer against
threat, even if it means disrupting the interpersonal transaction. Ir- the potential bias toward under-estimating cues of positive aect in
onically, this may be the basis of a self-fullling prophecy: these rup- faces.
tures in the transaction may lead their interaction partner to pull away
in the very manner that was feared (Downey and Feldman, 1996). 4.6. Conclusions
However, eortful control protected against this response bias. The
capacity for self-regulation may be essential for translating the detec- Borderline personality features were associated with greater emo-
tion of social cues into positive, mutual relatedness. Good eortful tion sensitivity to facial expressions, i.e., with a greater ability to de-
control could prevent those with high borderline personality features tect and label facial emotions at low intensity thresholds, and a greater
from detecting subtle signs of emotions in ambiguous or even neutral propensity to misdetect and mislabel neutral faces as emotional.
faces that in actual social interactions would be best left ignored. This is However, good eortful control buered this response bias in the de-
consistent with the notion that the empathy paradox may be attri- tection phase of facial emotion recognition. Individuals with high
butable in part to a combination of enhanced attention/perception of borderline personality features may therefore be automatically noti-
social stimuli with dysfunctional regulatory systems; in BPD a poten- cing negative emotions in faces that are actually not or barely present,
tially pathological facial emotion recognition pattern may become ap- but good self-regulatory skills may protect them against this potential
parent only in the absence of ecient top-down regulatory systems social-cognitive vulnerability.
(i.e., Dinsdale and Crespi, 2013). Conversely, when self-regulatory
processes work properly, those with high borderline personality fea- References
tures may overcome their automatic tendencies toward amplifying
minor signs of negative aect in others, resulting in better social ad- Adolphs, R., 2002. Neural systems for recognizing emotion. Current opinion in neuro-
justment. biology. 12 (2), 169177.
Ayduk, O., Zayas, V., Downey, G., Cole, A.B., Shoda, Y., Mischel, W., 2008. Rejection
sensitivity and executive control: joint predictors of borderline personality features.
4.5. Limitations and future directions J. Res. Personal. 42, 151168.
Bland, A.R., Williams, C.A., Scharer, K., Manning, S., 2004. Emotion processing in bor-
derline personality disorder. Issues Ment. Health Nurs. 24, 655672.
This study employed a non-clinical sample mostly comprised of Benton, A.L., 1994. Contributions to Neuropsychological Assessment: A Clinical Manual.
female undergraduate students, thus calling for a replication of ndings Oxford University Press, USA.
among older, male, and clinical populations. However, the ethnic di- Cain, N.M., De Panlis, C., Meehan, K.B., Clarkin, J.F., 2013. Assessing interpersonal
proles associated with varying levels of eortful control. J. Personal. Asses. 95 (6),
versity in our sample increases the generalizability of the ndings be- 640644.
yond predominately Caucasian samples, and the range of borderline Chanen, A.M., Jovev, M., Djaja, D., McDougall, E., Yuen, H.P., Rawlings, D., Jackson,
personality pathology endorsed by the current study participants (as H.J., 2008. Screening for borderline personality disorder in outpatient youth. J.
Personal. Dis. 22 (4), 353364.
evidenced by a mean of 5 borderline personality features, with a range
Daros, A.R., Zakzanis, K.K., Ruocco, A.C., 2013. Facial emotion recognition in borderline
of 013 features on the SCID-II screener) allows for the evaluation of personality disorder. Psychol. Med. 43 (9), 19531963.
facial emotion recognition accuracy across a full dimension of border- Daros, A.R., Uliaszek, A.A., Ruocco, A.C., 2014. Perceptual biases in facial emotion re-
line personality features. Further, using a non-clinical sample has the cognition in borderline personality disorder. Personal. Disord. 5 (1), 7987.
De Panlis, C., Meehan, K.B., Cain, N.M., Clarkin, J.F., 2013. The relationship between
potential to illuminate specic facial emotion recognition patterns re- eortful control, current psychopathology, and interpersonal diculties in adult-
lated to BPD traits rather than other illness-related variables. For in- hood. Compr. Psychiatry 54, 454461.
stance, non-clinical individuals are less likely to exhibit the inter- De Panlis, C., Meehan, K.B., Cain, N.M., Clarkin, J.F., 2016. Eortful control, rejection
sensitivity, and borderline personality disorder features in adulthood. J. Personal.
personal and cognitive impairments that usually characterize patients Disord. 30 (5), 595612.
with BPD (i.e., due their prolonged illness duration, comorbidity and Dinsdale, N., Crespi, B.J., 2013. The borderline empathy paradox: evidence and con-
medication use). However, the present study is limited by the absence ceptual models for empathic enhancements in borderline personality disorders. J.
Personal. Disord. 27 (2), 172195.
of data on the full range of PD symptoms on the SCID-II screener, as Domes, G., Czieschneck, D., Weidler, F., Berger, C., Fast, K., Herpertz, S.C., 2008.
well as some missing data among available subjects. The present nd- Recognition of facial aect in borderline personality disorder. J. Personal. Disord. 22,
ings should be replicated with conrmed PD symptoms on the SCID-II 135147.
Domes, G., Schulze, L., Herpertz, S.C., 2009. Emotion recognition in borderline person-
Interview, both within nonclinical and clinical samples; with a cut-o ality disorder: A review of the literature. J. Personal. Disord. 23, 619.
of > 12 on the SCID-II questionnaire indicating the presence of BPD Downey, G., Feldman, S.I., 1996. Implications of rejection sensitivity for intimate re-
(Chanen et al., 2008), very few in this sample would likely have met full lationships. J. Personal. Soc. Psychol. 6, 13271343.
Dyck, M., Habel, U., Slodczyk, J., Schlummer, J., Backes, V., Schneider, F., Reske, M.,
diagnostic criteria. Utilizing the full SCID-II Interview would also allow
2009. Negative bias in fast emotion discrimination in borderline personality disorder.
for an evaluation of the full spectrum of PD pathology, which may in- Psychol. Med. 39 (05), 855864.
teract with borderline personality features in predicting facial emotion Ekman, P., Friesen, W., 1976. Pictures of Facial aect. Consulting Psychologist Press, Palo
recognition accuracy. Alto, CA.
Evans, D.E., Rothbart, M.K., 2007. Developing a model for adult temperament. J. Res.
We do not have information on comorbidities of our undergraduate Personal. 41, 868888.
sample, though comorbidity has generally not been borne out as a First, M.B., Spitzer, R.L., Gibbon, M., Williams, J.B.W., 1997. Structured Clinical
signicant contributor of facial emotion recognition in BPD samples Interview for DSM-IV Personality Disorders, (SCID-II). American Psychiatric Press
Inc, Washington, DC.
(Daros et al., 2014). However, facial recognition may be altered in Fenske, S., Lis, S., Liebke, L., Niedtfeld, I., Kirsch, P., Mier, D., 2015. Emotion recognition
depressive disorders (Jenness et al., 2015), and trauma has been found in borderline personality disorder: eects of emotional information on negative bias.
to relate to accuracy for full-intensity emotion in faces in those with Borderline Personal. Disord. Emot. Dysregul. 26 (2), 10.
Gardner, K.J., Qualter, P., Stylianou, M., Robinson, A.J., 2010. Facial aect recognition in
BPD (Lowyck et al., 2016). Trauma and dissociation have also been non-clinical adults with borderline personality features: The role of eortful control
found to interact with borderline personality features on the RMET and rejection sensitivity. Personal. Individual Dier. 49, 799804.
(Weinstein et al., 2016), so future research would benet from their Germans, S., Van Heck, G.L., Mastho, E.D., Trompenaars, F.J., Hodiamont, P.P., 2010.
Diagnostic eciency among psychiatric outpatients of a self-report version of a
inclusion in the evaluation of accuracy for low intensity emotions.

Los art?culos se proporcionan ?nicamente para uso personal y educativo. Est? prohibida la reproducci?n, distribuci?n y uso comercial.

K.B. Meehan et al. Psychiatry Research 255 (2017) 347354

subset of screen items of the structured clinical interview for DSMIVTR Personality borderline personality disorder. Compr. Psychiatry 47, 468474.
Disorders (SCID-II). Psychol. Assessment 22 (4), 945. Mitchell, A.E., Dickens, G.L., Picchioni, M.M., 2014. Facial emotion processing in bor-
Goodman, J.K., Cryder, C.E., Cheema, A., 2013. Data collection in a at world: The derline personality disorder: a systematic review and meta-analysis. Neuropsychol.
strengths and weaknesses of Mechanical Turk samples. J Behav. Decis. Making 26 (3), Rev. 24 (2), 166184.
213224. Phillips, L.H., Channon, S., Tunstall, M., Hedenstrom, A., Lyons, K., 2008. The role of
Guitart-Masip, M., Pascual, J.C., Carmona, S., Hoekzema, E., Berg, D., Prez, V., working memory in decoding emotions. Emotion 8 (2), 184191.
Vilarroya, O., 2009. Neural correlates of impaired emotional discrimination in bor- Piedmont, R.L., Sherman, M.F., Sherman, N.C., Williams, J.E.G., 2003. A First Look at the
derline personality disorder: an fMRI study. Prog. Neuro-Psychoph. 33 (8), Structured Clinical interview for DSM-IV Personality Disorders Screening
15371545. Questionnaire: More than Just a Screener. Measurement & Evaluation in
Hayes, A.F., 2013. Introduction to Mediation, Moderation and Conditional Process Counseling & Development 36 (3), 150160.
Analysis. A Regression-Based Approach. Guilford Press, New York. Posner, M.I., Rothbart, M.K., 2009. Toward a physical basis of attention and self-reg-
Hidalgo, N.A.I., Oelkers-Ax, R., Nagy, K., Mancke, F., Bohus, M., Herpertz, S.C., Bertsch, ulation. Phys. Life Rev. 6, 103120.
K., 2016. Time course of facial emotion processing in women with borderline per- Preti, E., Richetin, J., Suttora, C., Pisani, A., 2016. Individual dierences in components of
sonality disorder: an ERP study. J. Psychiatry Neuroscience 41 (1), 1626. impulsivity and eortful control moderate the relation between borderline person-
Hoermann, S., Clarkin, J.F., Hull, J.W., Levy, K.N., 2005. The construct of eortful ality disorder traits and emotion recognition in a sample of university students.
control: An approach to borderline personality disorder heterogeneity. Psychiatry Res. 238, 109115.
Psychopathology 38, 8286. Robin, M., Pham-Scottez, A., Curt, F., Dugre-Le Bigre, C., Speranza, M., Sapinho, D.,
Jenness, J.L., Hankin, B.L., Young, J.F., Gibb, B.E., 2015. Misclassication and identi- Kedia, G., 2012. Decreased sensitivity to facial emotions in adolescents with bor-
cation of emotional facial expressions in depressed youth: A preliminary study. J. derline personality disorder. Psychiatry Res. 200 (2), 417421.
Clinical Child Adol. Psychology 44 (4), 559565. Rothbart, M.K., Ellis, L.K., Posner, M.I., 2011. Temperament and Self-Regulation. In: In:
LaBar, K.S., Crupain, M.J., Voyvodic, J.T., Mccarthy, G., 2003. Dynamic perception of Vohs, K.D., Baumeister, R.F. (Eds.), Handbook of Self-Regulation: Research, Theory,
facial aect and identity in the human brain. Cereb. Cortex 13 (10), 10231033. and Applicationspp. 441460.
Levine, D., Marziali, E., Hood, J., 1997. Emotion processing in borderline personality Rothbart, M.K., Bates, J.E., 1998. Temperament. Handbook of Child Psychology.
disorders. J. Nerv. Ment. Dis. 185 (4), 240246. Wiley, NY.
Lieb, K., Zanarini, M.C., Schmahl, C., Linehan, M.M., Bohus, M., 2004. Borderline per- Schulze, L., Domes, G., Koppen, D., Herpertz, S.C., 2013. Enhanced detection of emotional
sonality disorder. Lancet 364 (9432), 453461. facial expressions in borderline personality disorder. Psychopathology 46, 217224.
Lowyck, B., Luyten, P., Vanwalleghem, D., Vermote, R., Mayes, L.C., Crowley, M.J., 2016. Tottenham, N., Tanaka, J.W., Leon, A.C., McCarry, T., Nurse, M., Hare, T.A., Nelson, C.,
What's in a face? Mentalizing in borderline personality disorder based on dynamically 2009. The NimStim set of facial expressions: judgments from untrained research
changing facial expressions. Personal. Disord 7 (1), 7279. participants. Psychiatry Res. 168 (3), 242249.
Lynch, T.R., Rosenthal, M.Z., Kosson, D.S., Cheavens, J.S., Lejuez, C.W., Blair, R.J., 2006. Unoka, Z., Fogd, D., Fuzy, M., Csukly, G., 2011. Misreading the facial signs: specic
Heightened sensitivity to facial expressions of emotions in borderline personality impairments and error patterns in recognition of facial emotions with negative va-
disorder. Emotion 6, 647655. lence in borderline personality disorder. Psychiatry Res. 189, 419425.
Matsumoto, D., Ekman, P., 1988. Japanese and Caucasian Facial Expressions of Emotion Veague, H.B., Hooley, J.M., 2014. Enhanced sensitivity and response bias for male anger
(JACFEE). Intercultural and Emotion Research Laboratory. Department of in women with borderline personality disorder. Psychiatry Res. 215 (3), 687693.
Psychology, San Francisco State University, San Francisco. Wagner, A.W., Linehan, M.M., 1999. Facial expression recognition ability among women
Merkl, A., Ammelburg, N., Aust, S., Roepke, S., Reinecker, H., Trahms, L., Sander, T., with borderline personality disorder: Implications for emotion regulation? J.
2010. Processing of visual stimuli in borderline personality disorder: a combined Personal. Dis. 13, 329344.
behavioural and magnetoencephalographic study. Int. J. Psychophysiol. 78 (3), Weinstein, S.R., Meehan, K.B., Cain, N.M., Ripoll, L.H., Boussi, A.R., Papouchis, N., New,
257264. A.S., Siever, L.J., 2016. Mental state identication, borderline pathology, and the
Minzenberg, M.J., Poole, J.H., Vinogradov, S., 2006. Social emotion recognition in neglected role of childhood trauma. Personal. Disord. 7 (1), 6171.