You are on page 1of 5

PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 54th ANNUAL MEETING - 2010 487

Politeness Effects in Directive Compliance:


Effects with Power and Social Distance
Christopher A. Miller, Tammy Ott, Peggy Wu, Vanessa Vakili
Smart Information Flow Technologies

We present a theory of perceived politeness and its sociological functions derived from the work of Brown
and Levinson (1987) and then extend that theory toward a cognitive model of politeness and its effects on
human decision making. We then report the results of an experiment in which participants directive
compliance behaviors and attitudes are examined under conditions varying the amount of politeness or
rudeness used and the power or familiarity relationship between the participant and the directive giver.
Results show significant impacts of politeness on a variety of directive compliance behaviors, and show
accuracy for predicting the relationship of Social Distance on perceived politeness and directive com-
pliance. Predictions about the role of Power relationships were generally not as effective.

INTRODUCTION guages. Based on that corpus, they proposed that the cultu-
Politeness is seldom studied in the field of human factors. rally-universal function of politeness is to redress face threat.
This may be due to a historical focus on designed artifacts-- That is, any action I take with regard to another human has
machines and equipmentwhich may be assumed to not par- the potential to intrude on his or her autonomy of thought and
ticipate in social interaction dimensions normally reserved for action. If I take the action baldlythat is, without any
humans. Two trends make this assumption increasingly less form of mitigation, apology, or redressthen I may be im-
tenable: (1) machines themselves are getting more complex plying that I have the right or power to make that intrusion.
and designers are increasingly seeking to give them human- What we typically regard as politeness behaviorsthe use of
like qualities (e.g., Breazeal, 2002; Cassell, 2000), and (2) please, thank you, honorifics, etc.are redressive acts
there is increasing evidence that we frequently interact with that we use to offset the face threat inherent in interaction
Copyright 2010 by Human Factors and Ergonomics Society, Inc. All rights reserved. 10.1518/107118110X12829369603128

complex machines using the same expectations and interpre- whenever we dont wish to convey such a message.
tations that are afforded to other humans (Nass, et al., 1997). Thus, the degree of face threat present is a critical deter-
Politeness plays a role in many issues central to human miner of how an interaction will be perceivedand of how
factors: trust, team coordination, perceived workload, etc. much redress is required to offset it. Brown and Levinson
Indeed, the literature suggests that interaction style affects propose three factors as influencing face threat:
team performance (McNeese, Salas & Endsley, 2001) and 1. Power Difference that the Hearer has over the Speaker.
that effect influences trust (Lee & See, 2004). Cockpit Re- A less powerful individual will threaten face simply by
source Management (Weiner, Kanki & Helmreich, 1993), addressing a more powerful one. All other things be-
which has revolutionized team interactions in aviation and ing equal, if I am asking a favor of a peer, I can use
medicine can, in part, be regarded as both training operators less redress than of a boss or supervisor.
to be sensitive to personal style differences and, occasionally 2. Social Distance between the Hearer and Speaker. So-
reducing those differences through defined protocols for criti- cial distance is roughly the inverse of familiarity. Fa-
cal interactions. Finally, our own work (Parasuraman & Mil- miliar individuals (co-workers, family members,
ler, 2004) suggests that politeness can make profound per- friends) are expected to address one another, thus fami-
formance differences in human-machine systems. liarity reduces face threat. I can ask a favor of a friend
We have recently developed a functional model of po- using less redress than of a stranger.
liteness emphasizing how and why actions are perceived as 3. Imposition of the request or topic. Some topics are
polite, and how polite behaviors can alter social contexts. simply more imposing or face threatening than others.
This work stems from the sociolinguistic studies of Brown All other things equal, if I am asking a small favor, I
and Levinson (1987) which will be described next, followed can use less redress than if I am asking for a large one.
by a description of our extensions to that model to predict These factors drive face threat up and require more, or
behaviors and attitudes associated with perceived politeness. more powerful, redressive acts to reduce it. Inherent in
Finally, we present experimental results illustrating politeness Brown and Levinson is the assumption, which we have made
effects in perception of and compliance with directives. explicit and computational (Miller, et al, 2007; Miller, Wu &
Funk, 2008), that using enough redressive value to balance
A SOCIOLINGUISTIC MODEL OF POLITENESS face threat yields an assessment of nominal politeness in the
PERCEPTIONS AND FUNCTIONS observerabout as much politeness as might be expected for
this situation. The use of more politeness than the observer
Brown and Levinson (1987) collected a large corpus of thought necessary yields an assessment of over-politeness,
instances of politeness usage from multiple cultures and lan- while using less results in an assessment of rudeness.
PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 54th ANNUAL MEETING - 2010 488

Note that this model offers an explanation for an ob- rives from the likely increase in trust and positive affect
served property of politeness behaviors: that the same beha- that comes with expected, pleasing and/or more than
vior can be only nomimally polite in some contexts, far too nominally polite interactions. Intuitively, we certainly
polite in others, and not polite enough in still others. In behave as if politenesss will have this effecttraining
Brown and Levinsons terms, this is because elements of the our children to say please when asking for something if
context differ (as captured by the Power Difference, Social they want it-- and some specific experimental data are
Distance and Imposition terms) and, hence, demand that provided in (Parasuraman and Miller, 2004) to support
more or less redress be used to balance the resulting threat. the effectiveness of politeness. On the other hand, the
Brown and Levinson take this one further step by indicat- model itself suggests that the relationship is more com-
ing that perceptions of the face threat parameters can them- plex: I may choose to comply because I like you (a feel-
selves be influenced by the amount of politeness used. If I use ing that will be enhanced by reduced social distance), be-
less politeness than you thought was necessary for a given cause I fear you (which may be enhanced by increased
interaction, you might perceive me as simply rude but you power), because I feel it is in my best interests to do so
might also seek to balance the equation by adjusting your (low Imposition to me) or because it costs me little to do
prior perception of the Power Difference, Social Distance or so (low Imposition overall). The use of redressive beha-
Imposition of the context. You may have thought I was rude viors can steer the Hearer toward specific interpretations
because you thought the person I was addressing had power based on prior assumed relationships and complex issues
over mebut my relative rudeness suggests that perhaps I of personality and motivation. Nevertheless, in the ab-
think I have power over him/her, or perhaps we are old sence of such special cases, it seems reasonable to predict
friends, or perhaps what we are talking about is not as impos- that increased politeness should enhance compliance.
ing as you assumed, etc. Perceived Workload and Memorability: A cognitive in-
terpretation of the Brown and Levinson model suggests
POLITENESS AND DIRECTIVE COMPLIANCE that deviation from expected (nominal) levels of polite-
ness provokes increased reasoning about the interaction
Building on the Brown and Levinson model described and its context. Thus, we would expect off-nominal in-
above, we would like to project the effects of perceived polite- teractions to be associated with higher cognitive work-
ness on directive compliance and associated behaviors and load as the hearer tries to decipher possible hidden mes-
attitudes such as reaction time, memory, affect, trust, and sages. Similarly, memorability (memory for the inte-
perceived workloadwhich we will collectively refer to as raction and its social context, rather than overall situa-
directive compliance behaviors. Intuition and elements of the tion awareness) might be expected to improve under off-
existing literature led us (via a method described in more de- nominal circumstances where the hearer spends addi-
tail in Miller & Smith, 2008) to the following predictions: tional time scrutinizing initial assumptions.
Trust and Affect: We hypothesize that both trust in a di-
rective giver and affect about the situation will improve A TESTBED FOR POLITENESS EVALUATIONS
as the perceived politeness increases, while perceived
rudeness will decrease trust and affect. This will hold To begin testing the above hypotheses, we created a
true within a boundary of believable levels of politeness testbed which enabled us to vary and control aspects of the
and rudenessas politeness becomes unbelievable, more context which the models tell us are of interest. The result
attention is paid to interpretation and motives of the was the Park Asset Management and Monitoring Interface
speaker and trust and affect will decline. Relevant results (PAMMI) testbed which built on the Tactical Tomahawk In-
for the relationship between trust and affect are summa- terface for Monitoring and Retargeting (TTIMR-- Cummings
rized by Lee and See, (2004), by Norman (2002) for the and Guerlain, 2004). Although TTIMR was created to study
relationship between pleasure and affect and by Cialdini, UI impacts on users ability to control multiple unmanned
(1993) for the relationship between flattery and affect. vehicles, we were interested in the use of the chat channel
Reaction Time: Brown and Levinson themselves suggest interface as a means of both giving directives to users and of
(pp. 95-96) that one context in which reduced redress is controlling the contextual variables important to the interpre-
permissible is when action is both urgent and in the tation of perceived politeness in those directives.
hearers interests. Perceived rudeness may thus result in PAMMI presented a map illustrating the planned and ac-
shorter reaction times because it suggests urgency, while tual paths of multiple air and ground vehicles heading to dif-
politeness conveys reduced urgency. This effect may be ferent destinations in a National Park, notionally to fight a
highly sensitive: as more rudeness is used, net reaction forest fire. Additionally, a matrix of vehicles names (in rows)
times might increase as the hearer spends more time and destination names (in columns) showed which vehicles
wondering why the speaker is behaving so rudely. were heading to which destinations, along with their pro-
Compliance: We hypothesize a general (but not uni- jected arrival times (in the tables cells). A timeline offered a
form) increase in overt compliance with perceived polite- relational view of vehicle mission and arrival times to aid in
ness, at least within the believability window. This de- answering questions such as which vehicle will arrive next?
PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 54th ANNUAL MEETING - 2010 489

that DGs square was ungrayed and the message appeared


in the text box. Participants could read the message, close the
incoming message screen, return to the main interface to de-
termine the answer to the information request, and then select
an Outgoing Message screen to input their answer.
The information content of each directive was randomly
varied across DGs, but each of the 5 DGs was consistent in
their use of polite, rude or nominal directives: 2 were consis-
tently polite, 2 rude, and 1 nominal. In addition, we varied
the relationship of the participant to the DGs along power or
familiarity dimensions as will be described below. In order to
aid subjects in remembering which DG occupied which rela-
tionship, DGs were named and given icons which reinforced
both their individual identity and their role.
Figure 1. PAMMI's Incoming Message screen. Icons and text were used to indicate the DGs( rather than
and will vehicle A arrive before or after vehicle B? Finally, photos and voice or live video) to reduce age, sex, and cultur-
a chat window gave summary records of incoming and out- al associations, and so that the tone of voice would not inter-
going messages, though a larger popup window presented fere with the designed level of politeness. This attenuated the
messages along with indications of who they came from. range of cues available for inferring relationships (e.g., power
Although the TTIMR testbed required participants to and familiarity) but allowed us to control variability in those
control and direct vehicle movements, we eliminated this task interpretations. We suspect that such restricted interactions
in favor of placing participants in the role of a dispatcher also reduce the effects of perceived politenessthus making
who, through better access to information, simply observed this a conservative test of perceived politeness effects.
and reported on vehicle behaviors but did not influence them Incoming message events occurred once per minute and
directly. This was done so we could avoid having the poten- trials lasted 45 minutes after a period of training. Of the 45
tially unpredictable state of the simulated world provide un- incoming message events, 25 were single directives (one
predictable variations in the imposition that a directive would field agent making a request) and 20 were paired directives
produce. A directive requesting or ordering information (two field agents simultaneously issuing directives). Single
available from the participants screens could be expected to directives were used to measure reaction time; paired direc-
always produce about the same level of raw imposition. tives were used to present a forced choice and were our pri-
Participants were told they would support a group of five mary measure of compliance rates. Reaction times were also
field agents who would periodically ask for information. collected for paired directives, but were confounded by partic-
Directives for information arrived via dialog screen showing ipants potentially differing reading and selection strategies.
the requestors icon and a text message (cf. Figure 1). Direc- Accuracy rates and memorability were also assessed for both
tives consisted of an information request which could be ran- types of directives, and subjective ratings of affect, trust and
domly combined with politeness behaviors drawn from either perceived workload were assessed in a posttest.
a polite, nominal or rude group as determined by our use of
the Brown and Levinson theory. Thus, for example, the in- EXPERIMENTS WITH POLITENESS DIMENSIONS
formation request the arrival time of UTRUCK 018?
could be combined with the polite prefix Could you please We will report the results of two experiments testing our
let me know, the nominal prefix Tell me or the rude hypotheses about the effects of politeness on directive com-
prefix Stop being lazy and give me. All directives re- pliance. Each of these experiments used the PAMMI testbed
quired short, one-word answers, and participants were en- described above, but each varied one of the dimensions
couraged to answer as briefly as possible. Brown and Levinson had said should influence perceptions of
The set of directives were held constant across our expe- face threat and, therefore, of perceived politeness.
riments, but the set of directive givers (DGs) were varied. In
each experiment, 5 DGs were used and each occupied one Approach and Participants
square in the Incoming Message display illustrated in Fig- In Experiment 1, we sought to explore the effects of vari-
ure 1. Although the square occupied by each DG was varied ations in Power Difference on perceived politeness and on
across participants, it was held constant for that DG within a compliance behaviors. In Experiment 2, we explored varia-
trial. When an incoming message was from a specific DG, tions in Social Distance. In each, the general approach was
as outlined in the prior section with the following variations.
PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 54th ANNUAL MEETING - 2010 490

To vary these social context dimensions, we Politeness Ratings Compliance Rates


introduced a back story during training and rein- Exp 1
forced it in icon design. In Experiment 1, to vary
Exp 1
Hi Pol Lo Pol Hi Pol Lo Pol
power relationships, participants were told about Hi Pow 8.00 3.11 Hi Pow 0.83 0.79
the organizational hierarchy: that, as a dispatcher, Lo Pow 7.67 2.58 Lo Pow 0.29 0.23
they were subordinate to park commanders, but
superior to park rangers. This relationship was Exp 2 Exp 2
reinforced by an organization chart showing these Hi Pol Lo Pol Hi Pol Lo Pol
relationships included in the training materials and Hi Fam (Lo SD) 8.60 4.35 Hi Fam (Lo SD) 0.78 0.42
posted prominently during the trial. Finally, a Lo Fam (Hi SD) 6.75 2.35 Lo Fam (Hi SD) 0.50 0.18
Table 1. Mean politeness ratings (left column) and compliance rates (right column) of DGs
number of stars (1-3) presented next to the DGs by power level (Exp 1) or Social Distance level (Exp 2) and by Politeness used.
icon (cf. Figure 1) indicated the rank of that DG.
Similar manipulations were used in Experiment 2 to convey umn of Table 1. Even when using the same redressive beha-
team membership to vary social distancefor example, par- viors, socially distant speakers (those from a different team
ticipants were given a physical badge which, through colors and park) were rated less polite than near ones: F(1,19)=8.4,
and animal icons (Team Bird), signified the same team affili- p<.01. No significant effect was observed for Power.
ation with some of the DGs (bird icons in dark green), close Objective Compliance Effects. Compliance results are
affiliation with another team from the same park (team presented as a proportion of times the DG could have been
mammalin light green) and more distant affiliation (Team complied with (in paired directive events) on the right col-
Insectin white) with others from another park. umn of Table 1. In Experiment 1, we observed a significant
Of the five DGs each participant interacted with, each main effect of Power Difference, F(1,18)=39.30, p<.001, with
always used the same name and icon, and messages from that High Power DGs being complied with more frequently than
DG always appeared in the same location in the grid. The Low ones, but no effect or interaction with Politeness. In
five DGs fell into the following categories: Experiment 2, by contrast, social distance and politeness were
Two were high on the context dimension: higher pow- both significant, F(1,19)= 15.22 and 23.27, respectively, both
er than the participant or high social distance (unfami- p<.001. Familiar (Low SD) DGs were complied with more
liar, from the outside team). than Unfamiliar (High SD) ones. Polite DGs were complied
o Of these, one consistently used high politeness and with more than rude DGs.
the other consistently used low (i.e., was rude). No main effects or interactions were significant for reac-
tion time for single directive events, though a trend seemed to
Two were low on the context dimension: lower power
exist. In both Experiments 1 and 2, participants were slower
or low social distance (familiar, from the same team).
to respond to polite DGs than rude oneswhich was as we
o Of these, one consistently used high politeness and the
predictedthough this effect was small (M=38.3 sec for po-
other consistently used low politeness.
lite directives, 37.9 sec for rude ones).
One was neutral on the context dimension (same power
Although participants could clearly remember which di-
or same park but different team).
rective giver been more or less polite, no other memorability
o This neutral DG always used nominal politeness levels.
effects were found for either experiment. Because these ef-
Participants were selected from two Midwestern U.S.
fects were assessed via a series of posttest questions about
universities. We recruited heavily from international stu-
which DG had asked for which piece of information, it seems
dents organizations (due to interest in cultural differences),
likely that these simply proved too difficult. Correct responses
but the largest block of participants remained U.S. citizens.
were near chance levels for all experiments and conditions.
19 participants completed the first experiment; 20 the second.
Subjective Compliance Effects. Subjective effects of po-
liteness on affect, trust and perceived workload all behaved
Results
similarly. Politeness improved reactions as follows:
Politeness Assessments. Post test questions showed that
Polite DGs were rated significantly more likable than
our intended manipulations of politeness were effective. Par-
rude ones: Experiment 1F(1,21)=29.79, p<.001; Expe-
ticipants noticed and remembered which characters had been
riment 2F(1,19)=26.08, p<.001.
intended to be polite, nominal or rude (M= 7.8, 6.2 and 3.1
respectively) and DGs intended to be polite were rated as sig- Participants said they would trust the advice of polite
nificantly more polite in both experiments: F(1,21)=22.65, DGs more than rude: Experiment 1F(1,21)=16.04,
p<.001; F(1,19)=34.91, p<.001. p<.001; Experiment 2F(1,19)=26.75, p<.001.
An implication of Brown and Levinson is that increased Participants also trusted the competence of polite DGs
power of the speaker (relative to a constant hearer and utter- more than rude ones: Experiment 1F(1,21)=4.51,
ance) should increase perceived politeness, while increased p<.05; Experiment 2F(1,19)=9.81, p<.01.
social distance should decrease it. This effect was born out Participants generally reported lower perceived workload
for Social Distance but not for Power Difference in our expe- with polite DGs than with rude ones: Experiment 1--
rimentsas indicated by the politeness ratings in the left col- F(1,21)=4.54, p<.05; Experiment 2-- no significant ef-
PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 54th ANNUAL MEETING - 2010 491

fects on perceived workload were found, although the which do so as well. It may well be that compliance with
same trend was present in the data. power figures is a strong enough goal (especially among col-
Subjective effects also demonstrated our predicted out- lege students) in its own right, that the politeness or rudeness
comes for Social Distance, but not for Power Difference. of those individuals makes little difference in compliance.
Brown and Levinson predict that increased Power Difference On the other hand, note that Power Difference had little im-
(of Hearer over Speaker) should increase perceived politeness pact on perceived politeness. While powerful DGs were com-
of a constant utterance, and we predicted decreased likability, plied with more than weaker ones, even if they were rude,
trust and increased workload from that. Increased Social Dis- their rudeness did have a negative impact on how well they
tance was predicted to operate in reverse. We observed: were liked, trusted, how much workload participants per-
A significant effect of Social Distance on likability rat- ceived and (non-significant) how quickly they were reacted
ings in Experiment 2, F(1,19)=7.11, p<.05, such that fa- to. In short, this work clearly shows that politeness has im-
miliar DGs were rated more likable than unfamiliar ones portant impacts on directive compliance behaviors and atti-
regardless of politeness. By contrast, the prediction that tudes, even though it only begins to articulate and sort out the
increased Power Difference should increase perceived po- dimensions of those relationships. While our paradigm ex-
liteness and therefore likability was not born out. amined these effects in simulated human interactions, we
A significant effect of Social Distance (F(1,19)=20.40, claim that politeness considerations are and should be increa-
p<.001) such that trust in the advice provided by unfami- singly important in human factors engineeringboth because
liar DGs was rated lower than that from familiar ones. human team performance and cross cultural interactions are
No significant effects were found for Power Difference. increasingly a concern in human-technology interactions, and
A significant effect of Social Distance (F(1,19)=9.81, because human-technology interactions are increasingly acti-
p<.01) such that competence of unfamiliar DGs was vating and influenced by human social behaviors.
rated lower for familiar ones, regardless of politeness.
No significant effects were found for Power Difference. ACKNOWLEDGEMENTS
Neither Power Difference nor Social Distance produced a This work was supported by an SBIR grant (# FA8650-06-C-
significant effect on perceived workload. 6635) from the Air Force Research Laboratory. We thank
Ms. Kellie Plummer and Dr. Rik Warren, our Contract Moni-
CONCLUSIONS tors. We also acknowledge Curtis Hammond, Marie Kirsch,
Our data demonstrates clearly that perceived politeness Michael Wade, and Harry Funk for their contributions.
affects whether a directive is complied with and how partici-
pants feel about the person issuing it. Generally, politeness REFERENCES
Breazeal, C. (2002). Designing sociable robots. Cambridge, MA: MIT Press.
improved compliance ratesup to 34% in Experiment 2.
Brown, P. & Levinson, S. (1987). Politeness: Some Universals in Language
Politeness also significantly improved the affect, trust, and Usage. Cambridge,UK; Cambridge Univ. Press.
perceived competence that participants had toward DGsand Cassell, J. (2000). Embodied conversational agents. Cambridge, MA: MIT.
decreased the workload they perceived in taking directives. Cialdini, R. (1993). Influence: Science and Practice. New York: Harper-
Collins.
Compliance rates, reaction times (non-significant), af- Cummings, M. L. (2004). The need for command and control instant message
fect, trust and perceived workload were all improved by the adaptive interfaces: Lessons learned from tactical tomahawk human-in-the-
use of polite directives. Furthermore, Social Distance varia- loop simulations. CyberPsychology & Behavior, 7, 653-661.
Lee, J. D. & See, K. A. (2004). Trust in computer technology: Designing for
tions in relationships operated as predicted by Brown and appropriate reliance. Human Factors. 46 (1), 50-80.
Levinsonwith familiarity enhancing perceived politeness McNeese, M., Salas, E. & Endsley, M. (2001). New Trends in Cooperative
and its effects (except for workload and reaction time). By Activities. Santa Monica, CA; HFES.
Miller, C., Wu, P., Funk, H., Johnson, L. & Viljalmsson, H. (2007). A compu-
contrast, no such effects were observed for Power Difference,
tational approach to etiquette and politeness: An etiquette engine for
in spite of (our interpretation of) Brown and Levinsons pre- cultural interaction training. In Proceedings of BRIMS07 (pp. 189-198).
dictions that increased power of Speaker relative to Hearer Orlando, FL: Simulation Interoperability Standards Organization.
should also enhance politeness effects. In one sense, this very Miller, C. and Smith, K. (2008). Culture, Politeness and Directive Compliance:
Does Saying Please Make a Difference? In Proceedings of the NATO
lack of effect is confirmatory evidence for the model. Our RTO Symposium HFM-142 on Adaptability in Coalition Teamwork. Co-
hypothesis was that Power Difference should enhance per- penhagen, April 21-23, 2008.
ceived politeness with associated effects on directive com- Miller, C., Wu, P. & Funk, H. (2008). A Computational Approach to Etiquette:
Operationalizing Brown and Levinsons Politeness Model. IEEE Intelli-
pliance. The fact that we failed to observe a power difference gent Systems, 23(4), 28-35.
effect on perceived politeness would lead us to predict no ef- Nass, C., Moon, Y., Morkes, J., Kim, E., & Fogg, B. (1997). Computers are
fects on directive compliancewhich is what was observed. social actors: A review of current research. In B. Friedman (Ed.), Human
As for the broader question of why Social Distance might Values and the Design of Computer Technology. Cambridge University
Press, New York NY. 137-162.
interact with politeness to affect directive compliance, while Norman, D. (2002). Emotion & Design: Attractive things work better. Interac-
Power Differences seem largely independent of politeness, we tions Magazine, 9 (4). 36-42.
can offer only initial suggestions. Our simple model implies Parasuraman, R. & Miller, C. (2004). Trust and Etiquette in High-Criticality
Automated Systems. Communications of the ACM, 47(4), 51-55.
that, while perceived politeness may influence decision mak- Weiner, E., Kanki, B., & Helmreich, R. (1993). Cockpit resource manage-
ing, there are certainly other attitudes, beliefs and goals ment. San Diego: Academic Press.

You might also like