How Successful I am Depends on What Number I Get: The Effects of ...

AAPOR - ASA Section on Survey Research Methods
How Successful I am Depends on What Number I Get: The Effects of Numerical Scale
Labels and Need for Cognition on Survey Responses
Ting Yan
Survey Research Center, University of Michigan

Acknowledgment. The work reported here was part of
by Schwarz and his colleagues, attributes this type of
my Ph.D. dissertation submitted to the Graduate School,
measurement error to respondents’ use of Grice’s
University of Maryland. I am especially grateful to my
Cooperative Principle (CP) and its associated maxims in
academic advisor Roger Tourangeau and my committee
the survey research setting as in everyday conversation
members for their help and advice. I also like to thank
(see Schwarz 1996; for a review, see Tourangeau, Rips,
Norbert Schwarz and Frauke Kreuter for their and Rasinski 2000); the term “Gricean effect” is used
comments on an earlier draft.
throughout this paper to refer to this type of

measurement error. According to Grice, participants in
Abstract
communication are cooperative and rational; they speak
To further the understanding on the mechanism
in a truthful, informative, relevant, and clear manner
of numerical scale values, this study varied the (Grice 1989). Relying on this cooperative principle,
numerical labels for scale points and examined its effect
respondents make use of various visual features of
in relation to individual respondent’s need for cognition.
survey questions in interpreting questions and forming
Embedded in a web survey, this experiment succeeded
responses; they see the visual features as task elements
in producing a variety of evidence that respondents that are essential to the question-answer process (see
worked out an inference from the numerical values and
Couper, Tourangeau, and Kenyon 2004, for the
based their responses on that inference. The shift in
distinction between style elements and task elements).
responses induced by the numerical scale values was

The numerical values assigned to a rating scale
unexpectedly robust; when a scale started with a are one feature that seems to be taken as a task element
negative number, it pushed the responses to the right or
in the survey response process. Schwarz, Knauper,
the positive end of the scale across items and across
Hippler, Noelle-Neumann, and Clark (1991) first
fonts. Process measures (such as recall) and answers to
demonstrated that respondents came to different
the retrospective probe confirmed that respondents paid
interpretations of the verbal end points of a scale when
attention to the numerical labels on the scales and used
the scale ran from 0 to 10 than when it ran from –5 to 5.
them to interpret the verbal labels on the end points of
Presuming that every piece of information was relevant
the scale.
based on the Gicean Cooperative Principle, respondents
However, not everyone was affected by the inferred that the same end label (“not at all successful”)
numbers on the scales. The hypothesized effect of the
meant the “mere absence of noteworthy success” when
negative scale values was observed only among 0 was assigned to that scale point, but “the presence of
respondents with a high need for cognition, but not failure” when –5 was assigned to that point (Schwarz et
among those with a low need for cognition. This al. 1991). As a result, respondents were less likely to
finding seemed to suggest two things. First, Gricean
select values less than or equal to the midpoint with -5
effects of this sort involve controlled processes; people
to 5 scale labels than with the 0 to 10 labels; thus,
need to process deeply for the numbers to affect the
responses were pushed to the right or positive side of
answers. Second, unlike the response errors committed
the scale.
by satisficing respondents (who skip or slack off on
In addition to rating scales, Schwarz and his
certain cognitive steps), Gricean effects are an colleagues also investigated the effects of different
optimizing error or high effort error, committed by numerical values attached to a frequency scale (Schwarz,
respondents who try to be good, cooperative, and Grayson, and Knauper 1998, Experiment 1). The
thorough respondents.
frequency scale ranged either from 0 to 10 or from 1 to

11. The end labels remained “rarely” for 0 or 1 and
Keywords: Gricean effect, pragmatic inference, “often” for 10 or 11. Again, the numerical values of the
optimizing, satisficing, measurement error, web surveys
scales influenced the responses – respondents reported
higher frequencies when the scale ranged from 0 to 10
Introduction
than when it ranged from 1 to 11. Schwarz and
Prior studies have established that many colleagues speculated that the end label “rarely”
incidental features of survey questions can affect indicated a lower frequency when combined with value
respondents’ answers, creating unwanted response 0 than with value 1; as a result, the scale running from 0
errors (Schwarz 1999; 1996). One line of research, led

4262

---

AAPOR - ASA Section on Survey Research Methods
to 10 shifted the means to the higher end of the evaluation (Petty and Jarvis 1996; see Cacioppo, Petty,
frequency (Schwarz et al. 1998).
Feinstein, and Jarvis 1996 for a comprehensive review
This effect of numerical labels is replicated on
on need for cognition). The need for cognition can be
questions about different topics (Haddock and Carrick
considered a person’s tendency to think and enjoy the
1999). The numerical values assigned to scale points
process of thinking; thus, it reflects a stable personality
were shown to consistently affect survey responses trait rather than a temporary choice of cognitive strategy.
(regardless of the types of the scales used). The effect is
It is typically assessed with an 18-item need for
robust across modes of administration (telephone cognition scale created by Cacioppo and colleagues
interview vs. self-administered questionnaires vs. face-
(Cacioppo, Petty, and Kao 1984).
to-face interviews), for both unipolar and bipolar scales,
In the survey context, Petty and Jarvis (1996)
for various domains, and for self and proxy reports predicted that the HNC group is expected to process
(O’Muircheartaigh et al. 1995; Schwarz et al. 1991; survey questions more carefully than the LNC group.
Schwarz and Hippler 1995). Nonetheless, two Accordingly, they speculated that the HNC group is
important issues have yet to be addressed to gain a fuller
more susceptible to high effort biases such as primacy
understanding of the mechanism of numerical scale effects and priming effects while low effort biases – e.g.,
values.
errors resulting from yea-saying to agree-disagree items
The first one is the lack of direct evidence that
(or acquiescence) or providing similar responses to a
the response changes induced by the numerical values
batch of questions (non-differentiation) – are produced
were due to respondents’ utilization of the maxim of
by the LNC group only (Petty and Jarvis 1996).
relation during the survey response process. In most of
Krosnick’s notion of satisficing distinguishes
the past studies, changes in response distributions were
two types of respondents based on their response styles:
observed and were speculated to be caused by “optimizers” respond carefully and thoughtfully
respondents’ interpretation of scale labels as a result of
whereas “satisficers” short cut their cognitive processes
their utilization of the maxim of relation. Only one
by either executing cognitive steps less completely or
experiment out of the eight documented sought direct
skipping certain cognitive steps (Krosnick 1991; 1999).
evidence of respondents’ interpretation of the scale Consequently, satisficers are more attracted to
labels through a follow-up question.
cognitively untaxing response behaviors such as giving
Second, the existing studies mainly looked at
“Don’t Know” responses, acquiescing answers, non-
the overall main effects of the numerical labels, differentiated responses, and picking the first seemingly
implicitly assuming that this particular type of Gricean
reasonable response option (Krosnick 1991; 1999;
effect triggered by the numerical labels of the scales is
Narayan and Krosnick 1996).
across-the-board among all respondents. However,
Unlike the need for cognition, satisficing
respondents differ in how they answer survey questions
reflects a respondent’s response strategy and the
and how much cognitive effort they exert in survey likelihood of satisficing is a function of respondents’
response process. If taking numerical labels into cognitive ability, motivation, and task difficulty
consideration when forming responses involved (Krosnick 1991, 1999). There is not yet one method to
additional cognitive effort (i.e., attention to and assess satisifcing but Krosnick (1991, 1999) provided a
processing of the numbers), there would be some list of covariates that are said to be associated with
respondents who were either unable to or unwilling to
satisficing. For instance, the need for cognition affects
expend the additional cognitive effort in processing the
the likelihood of satisficing through their impact on
numerical scale values. Therefore, existing studies respondent cognitive ability whereas fatigue and
might have covered up subgroup differences with regard
boredom show their influence through motivation. Type
to the effects of numerical labels by focusing only on
and structure of survey questions have to do with task
main effects.
difficulty.
Two closely related notions have to do with
Though different in conceptualization and
individual differences in exerting cognitive efforts. One
operationalization, both the need for cognition and
concept is people’s need for cognition, which represents
satisficing predict individual differences in the amount
“the extent to which people tend to engage in and enjoy
of cognitive effort exerted in survey response process.
effortful cognitive activity” (Petty and Jarvis 1996, Both predict that those respondents who exert less
p.221). According to Petty and Jarvis, people differ in
cognitive effort respond to the scale numerical values
the cognitive effort they exert in thought processes. differently from those who exert more effort.
Those with a high need for cognition (HNC) tend to
Despite the large number of research studies on
seek out more information and process information need for cognition reviewed in Cacioppo et al. (1996),
more carefully whereas the group with a low need for
need of cognition is not commonly considered in the
cognition (LNC) is more inclined to adopt simple or
survey context. There are only a handful of studies
cognitively untaxing strategies before making an looking at the effect of need of cognition on survey

4263

---

AAPOR - ASA Section on Survey Research Methods
responses. Bizer, Krosnick, Holbrook, Petty, Rucker,
to utilize the maxim of relation, showing larger effects
and Wheeler examined the moderating effect of need for
of negative scale numbers than people low in need for
cognition in the 1998 National Election Survey Pilot
cognition will.
Study (2002). They found out that people low in need
for cognition were less likely to enjoy the survey The Study
process and were slightly more likely to say “don’t
Overview. This experiment was embedded in a
know” when asked attitude questions (Bizer et al. 2002).
web survey conducted by MS Interactive. Survey
However, Fournier, Lyle, Cutler, and Soroka, using the
Sampling Inc. (SSI) selected the sample for this study
same dataset, failed to confirm the hypothesis that need
from its opt-in Web panel (Survey Spot) of over one
for cognition is related to susceptibility to opinion million persons who have signed up online to receive
change (2004). One reason for the weak effect of need
survey invitations. SSI selected 17,362 e-mail
for cognition in the two studies lies in the inadequate
addresses for this study and sent out e-mail messages
measure of need for cognition; only two items out of the
inviting the recipients to take part in “a study of
original 18-item scale were included in the pilot study.
attitudes and lifestyles.” The e-mail invitations included
McCabe and Brannon (2004) reported a partial
the web address (URL) for the survey web site and a
replication of the general-specific questions with an unique identification number (which prevented
emphasis on the impact of a joint lead-in on responses. respondents from completing the survey more than
Both Schwarz, Strack, and Mai (1991) and Tourangeau,
once). The survey ran from May 24 to June 2, 2005. Of
Rasinski, and Bradburn (1991) demonstrated that, when
the 17,362 invited to participate in the survey, 1,071
a specific question (e.g., relationship satisfaction) completed the entire survey (and 146 others got part
preceded a general question (e.g., satisfaction with life
way through) for a response rate (AAPOR [2000] RR1)
in general), the correlation between the two questions
of 6%. The questionnaire included questions on a range
was reduced in the presence of a joint lead-in because
of topics, most of them attitudinal. The 18-item need
respondents applied the maxim of quantity to avoid for cognition scale (Cacioppo, Petty, and Kao, 1984)
providing redundant information. McCabe and Brannon
was included in the last section, together with
found that only the HNC respondents displayed such an
demographic questions. This experiment came first in
attenuated correlation between the items, but not the
the questionnaire.
LNC respondents. Their finding suggested that the
Experimental manipulation. This experiment
conversational norm to avoid redundancy was not manipulated both the numerical values assigned to the
automatically applied in the survey context; only those
scale points (replicating the earlier studies) and the
with a high need for cognition seemed to apply the appearance of the scale values in a 2 (numerical labels:
maxim. This is the first empirical evidence for the 0 to 6 vs. -3 to 3) x 2 (appearance: normal font vs. faint
prediction that the HNC group is subject to high effort
font) factorial design.1 Table 1 displays the number of
bias – errors resulting from respondents’ optimizing completes per experimental condition.
behavior.
Target questions. The key target question is
This study aimed to fill the gap in existing the success item. I used the same question wording as
literature, fulfilling two goals: a) seeking direct in Schwarz et al. (1991). For replication purpose,
evidence that respondents use the maxim of relation
respondents were also asked to rate their moodiness,
when answering rating scales questions with numerical
their nervousness, and optimism along one of the four
labels, and b) demonstrating that the use of the maxim
randomly assigned scales. Respondents got the same
of relation in this case is a high effort bias committed
numerical labels for all four questions.
only by respondents with a high need for cognition
(HNC).
Three process measures and one follow-up
question are employed to provide direct evidence of
respondents’ utilization of the maxim of relation. The

corresponding hypotheses regarding the process 1The faint font version of the scales displayed the
measures are that a negative scale number will induce
numerical values assigned to scale points in a distinct
better recall, greater attention and usefulness rating by
font that was much fainter than the font used for the
respondents. The follow-up question solicits directly
question text and the verbal label. Such fonts are
respondents’ inference about the scale label; typically used in paper questionnaires for information
respondents are hypothesized to infer a presence of that is not intended for the respondents. The purpose of
negative trait when the scale starts with a negative this font manipulation is to test whether a faint font
number. In addition, given the differential cognitive
would lead respondents to discount the relevance of the
effort exerted in the question-answer process, scale labels in a web survey. The analysis of the font
respondents high in need for cognition are hypothesized
manipulation is not presented in this paper.

4264

---

AAPOR - ASA Section on Survey Research Methods
Follow-up questions. The follow-up questions
If conversational implicatures were worked out, the
asked respondents about their use of the scale values
extra effort needed to work out an implicature should
and the inferences they drew about the scale end labels.2 produce better recall of the numbers that triggered the
interpretative maxim. Respondents should have also
Results
paid more attention to the numbers and considered the
I begin by presenting the analyses of responses
numerical values more useful when they used them in
to the four target questions, followed by analyses on
interpreting the response scale.
respondents’ use of and inferences about the scale.
I examined the percentage of respondents who
Responses. For all four scale conditions, I recalled the leftmost scale value correctly by the scale
coded the responses from 1 to 7, where 1 corresponded
values and need for cognition. Figure 2 indicates that,
either to 0 or -3, and 7 to 6 or 3. To compare responses
in general, more respondents recalled the number
to scales with different numerical values, I examined the
correctly when presented with the -3 to 3 scale labels
mean ratings of the 0 to 6 scales and of the -3 to 3
than with the 0 to 6 labels (χ2=3.18, p=0.07). However,
condition (see Table 2).
the HNC respondents were more likely to recall
As evident from Table 2, the negative scale
correctly the number assigned to the leftmost scale point
values produced a mean shift in the key target item
than the LNCs; the difference between the two
(success), replicating the finding by Schwarz et al. respondent groups was bigger when the scale started
(1991). In addition, mean shifts were replicated on the
with -3 rather than 0 (see Figure 2). The difference in
other three items too. One-way ANOVAs, conducted on
the percentage of correct recall by numerical label
all four target questions, confirmed that numerical conditions is marginally significant for the HNC group
values had a significant effect on responses for the first
(χ2=3.53, p=.06), but not for the LNC group (χ2=0.56,
three items, but not on the last one.
ns). Still, the pattern is consistent with the prediction –
To examine the effects of an individual’s need
the negative values were recalled better overall and the
for cognition on responses, I split the sample into two
difference in recall is more marked with the HNC group.
groups based on their scores on the need for cognition
Another two follow-up items asked
scale. Those who scored higher than 3.5 (median value)
respondents how much attention they paid to the
are considered to have a high need for cognition (HNCs) numerical values attached to the scale and how useful
whereas those with a score of 3.5 or lower are regarded
they considered those numbers. Figure 3 plots the
as the LNC group. Figure 1 plots the mean responses to
average ratings of attention and usefulness for the
the success item by the numerical labels and the need
numerical labels and the need for cognition,
for cognition. The negative scale number induced a
demonstrating that respondents tended to pay more
mean shift for both groups; however, consistent to my
attention and consider the scale label more useful when
hypothesis, the shift was significantly bigger for the scale started with a negative number than with zero.
respondents with a high need for cognition than for The effect of numerical values is significant for the
those with a low need for cognition.3 The main effects
attention ratings (F(1,1066)=15.73, p<.0001), but not
of numerical labels (F(1,1045)=13.46, p<.00), of need
significant for the usefulness ratings.
for cognition (F(1,1045)=18.57, p<.00), and the
Compared to the LNC group, the HNC
interaction effect (F(1,1045)=4.79, p<.05) are highly respondents consistently claimed to have paid more
significant.
attention to the negative numerical label (simple main
Process measures. I assessed the inferences
effect of numerical label: F(1,1066)=13.37, p<.000) and
respondents drew based on their answers to follow-up
to have considered the negative number more usefully
questions. I first examined various process measures. (F(1,1059)=3.15, p<.10) when the scale ran from -3 to 3.
The differences in ratings between the two numerical

label conditions are more perceptible with the HNC
2 The exact wordings of the target questions and follow-
respondents than with the LNC group; the relevant
up questions are not presented, but can be requested
simple main effects of numerical labels are not
from the author.
significant for the LNC respondents.
3 Analyses of the other three items revealed a significant
Inferences. Process measures such as recall
three-way interaction (item*numerical label*need for
task, and self-reported attention level and usefulness of
cognition). One post-hoc explanation was the
numbers suggested that inferences were drawn in
confounding of the connotative meanings of the
response process. In order to seek direct evidence and
moodiness and nervousness items with need for
to determine the exact inference respondents drew from
cognition. The HNC respondents were less likely to
the numerical values, the last follow-up question asked
consider themselves as nervous or moody; as a result,
respondents what the scale label “not at all successful”
they tended to place them on the left side of the scale,
rather than the right side of the scale.

4265

---

AAPOR - ASA Section on Survey Research Methods
meant to them.4 There were six answer categories to
probe confirmed that respondents paid attention to the
this question (see bottom panel of Appendix for the
numerical labels on the scales, carefully processed the
exact wordings of the six answer categories). I negative numbers, and worked out inferences to
collapsed the answer categories into two groups – one
interpret the verbal labels on the end points of the scale.
group represents the absence of success and the other
Process measures also seemed to suggest that the
group the presence of failure. Figure 4 plots the processing of negative numerical labels is a controlled
percentage of people inferring “presence of failure” by
process (evidenced by better recall).
scale numerical labels and the need for cognition.
This study further showed that this Gricean

The result supports the conjecture of Schwarz
effect is a high-effort bias committed by optimizing
and his colleagues (1991) about how respondents respondents; the effect of negative scale numbers is
interpret the scales with different numerical labels. more marked with respondents who have a high need
Significantly more respondents interpreted the scale for cognition than those with a low need for cognition.
label “not at all successful” to mean the presence of
In other words, more thinking and deeper processing by
failure when the numerical labels ran from -3 to 3 than
respondents high in need for cognition made the effects
when they ran from 0 to 6 (χ2=4.80, p=.03), suggesting
of negative scale numbers bigger. This finding, together
that respondents did take the numerical values into with the results from process measures, showed that
consideration when they constructed their answers.
processing negative scale values is a controlled process.
However, it is again the HNC group who used
Consistent with the finding by McCabe and Brannon,
the numerical values in interpreting scale verbal labels;
this study provides further empirical evidence that the
significantly more HNC respondents drew the inference
resulting Gricean effect of negative scale values is a
of “presence of failure” when the scale started with -3
high effort bias committed by respondents who read too
(73%) than when the scale ran from 0 to 6 (63%) much into survey questions and contexts.
(χ2=4.97, p=.03). By contrast, about same percentage of
This study points to a few issues that merit
LNC respondents interpreted the scale label “not at all
survey researchers’ attention. First, research on survey
successful” to mean the presence of failure when the
measurement errors have been focusing on identifying
numerical labels ran from -3 to 3 (62%) than when they
and fixing response errors committed by respondents
ran from 0 to 6 (65%); the difference is not significant
who are cognitive misers and who haven’t put in as
controlling for the need for cognition (χ2=.88, ns). many cognitive effort as we desire (Krosnick 1991;
Figure 4 demonstrates that it is respondents with a high
1999). The implicit assumption held by most survey
need for cognition that drew inferences from the researchers is that more thinking and deeper processing
negative scale number, replicating the finding by is better than little thinking and shallow processing.
McCabe and Brannon (2004) and supporting the claim
However, whether this is true or not may depend on the
by Petty and Jarvis (1996) that the HNC respondents are
specific response effect. As this study showed, careful
more susceptible to high effort biases.
processing could lead to errors as much as the lack of
careful processing. Therefore, survey researchers
Discussion
should shift away from their traditional emphasis on
errors committed by satisficing respondents to those by
Several studies by Schwarz and his colleagues
optimizing respondents.
demonstrated that the numerical values assigned to the
Second, survey researchers should start
scale points affect the distribution of the responses. develop techniques for overcoming optimizing errors or
This study replicates and extends previous work by high-effort biases. Traditional techniques for
Schwarz and colleagues by manipulating the numerical
overcoming errors resulting from satisficing – such as
labels for scale points and examining the moderating
motivating respondents – might not effectively reduce
effect of need for cognition. The results showed that the
errors by optimizing respondents. At least, encouraging
mean shift in response to the right side of the rating
respondents to think more carefully about the question
scale induced by negative numerical labels was robust
and the negative scale numbers would only trigger the
across items.
controlled processes and increase the mean shift caused

This study also provides direct evidence that
by the negative scale numbers.
respondents draw inferences about the verbal labels of
Third, Schwarz suggested that the best way to
the scale points based on the Gricean maxim of relation.
reduce Gricen effect is for survey researchers to become
Process measures such as recall task, self-reported a cooperative communicator (Schwarz 1998; 2000).
attention to the scale numbers, and the retrospective
Survey researchers have the responsibility to

communicate to respondents what should be perceived
4 I present here the results based on the closed-end
as informative and what should not be. Thus, it is no
question. Analyses of the open-ended responses were
longer enough to simply pretest survey questionnaires at
similar and didn’t change the conclusions reported here.
the level of semantics and syntax; survey researchers

4266

---

AAPOR - ASA Section on Survey Research Methods
should also pretest survey instruments at the level of
McCabe, Amy E. and Laura A. Brannon. 2004.
pragmatics in order to prevent Gricean effects reported
Application of Conversational Norms to the
here from happening (Tourangeau, Rips, and Rasinski
Interpretation of Survey Results as a Function
2000).
of Participants’ Need for Cognition.” The
Last, given that people with different levels of
Journal of Psychology 138:91-4.
need for cognition are subject to different types of O’Muircheartaigh, Colm A., George D. Gaskell, and
measurement error, survey researchers should consider
Daniel B. Wright. 1995. “Weighing
including the need for cognition scale in their
Anchors: Verbal and Numeric Labels for
instrument; the inclusion of such a scale could shed
Response Scales.” Journal of Official Statistics
light on respondents cognitive processing and efforts. It
11:295-307.
could also be used as a covariate to be controlled for in
Petty, Richard E., and W. Blair G. Jarvis. 1996. “An
their statistical modeling and analysis.
Individual Differences Perspective on

Assessing Cognitive Processes.” In Answering
References
Questions: Methodology for Determining
Bizer, George, Jon Krosnick, Allyson Holbrook,
Cognitive and Communicative Processes in
Richard Petty, Derek Rucker & Christian
Survey Research, ed. Norbert Schwarz and
Wheeler. 2002. “The impact of personality on
Seymour Sudman, pp. 221-58. San Francisco:
political beliefs, attitudes, and behavior: Need
Jossey-Bass Publishers
for cognition and need to evaluate”, Paper
Schwarz, Norbert. 1996. Cognition and Communication:
presented at the annual meeting of the
Judgmental Biases, Research
American Political Science Association.

Methods, and the Logic of Conversation.
Cacioppo, John, Petty, R. E., Feinstein, J., & Blair
Mahwah, NJ: Erlbaum.
Jarvis. 1996. “Dispositional differences in
Schwarz, Norbert. 1998. “Warmer and More Social:

cognitive motivation: The life and times of
Recent Developments in Cognitive
individuals varying in need for cognition”,
Psychology.”
Annual Review of Sociology
Psychological Bulletin, 119: 197-253.
24:239-64.
Cacioppo, John T., Richard E. Petty, and Chuan Feng
Schwarz, Norbert. 1999. “Self-Reports: How the
Kao. 1984. “The Efficient Assessment
Questions Shape the Answers.” American

of Need for Cognition.” Journal of Personality
Psychologist 54: 93-105.
Assessment 48:306-07.
Schwarz, Norbert. 2000. “Social Judgment and
Couper, Michael. P., Roger Tourangeau, and Kristen
Attitudes: Warmer, More Social, and Less
Kenyon. 2004. “Picture this! Exploring
Conscious.”
European Journal of Social

Visual Effects in Web Surveys.” Public
Psychology 30: 149-76.
Opinion Quarterly 68:255-66.
Schwarz, Norbert, Grayson, Carla E., and Barbel
Fournier, Patrick, Greg Lyle, Fred Cutler, and Stuart
Knauper. 1998. “Formal Features of Rating
Soroka. 2004. “Need for Cognition,
Scales and the Interpretation of Question
Need to Evaluate, and Electoral Attitude
meaning.” International Journal of Public
Change,” Paper presented at the annual
Opinion Research 10:177-84.
meeting of the American Association for
Schwarz, Norber, and Hans-Jurgen Hippler. 1995. “The
Public Opinion Research.
Numeric Values of Rating Scales: A
Grice, H. Paul. 1989. Studies in the Way of Words.
Comparison of Their Impact in Mail Surveys
Harvard University Press.
and Telephone Surveys.” International Journal
Groves, Robert M. 1989. Survey Eerror and Survey
of Public Opinion Research 7:72-4.
Cost. New York: John Wiley and Sons.
Schwarz, Norbert, Barbel Knauper, Hans-Jurgen
Haddock, Geoffrey, and Rachael Carrick. 1999. “How
Hippler, Elizabeth Noelle-Neumann, &
to Make a Politician More Likeable
Clark, Leslie. (1991). “Rating Scales: Numeric
and Effective: Framing Political Judgment
Values May Change the Meaning of Scale
Through the Numeric Values of a Rating
Labels.” Public Opinion Quarterly 55: 570-82.
Scale.” Social Cognition 17:298-311.
Schwarz, Norbert, Fritz Strack, and Hans-Peter Mai.
Krosnick, Jon A. 1991. “Response Strategies for Coping
1991. “Assimilation and Contrast
With the Cognitive Demands of
Effects in Part-Whole Question Sequences: A

Attitude Measures in Surveys.” Applied
Conversational Logic Analysis.” Public
Cognitive Psychology 5:213-36.
Opinion Quarterly 55: 3-23.
Krosnick, Jon. A. 1999. “Survey Research.” Annual
Tourangeau, Roger, Kenneth A. Rasinski , and Norman
Review of Psychology 50: 537-67.
Bradburn. 1991. “Measuring

4267

---

AAPOR - ASA Section on Survey Research Methods
Happiness in Surveys: A Test of the
Tourangeau, Roger, Lance Rips, and Kenneth A.
Subtraction Hypothesis.” Public Opinion
Rasinski. 2000. The Psychology of Survey
Quarterly 55: 255-66.
Response. Cambridge University Press.

Table 1. Number of Completes Per Experimental Condition

0 to 6
-3 to 3
Total
Normal Font
259
271
530
Faint Font
271
270
541
Total 530
542
1071

Table 3. Mean Responses By Numerical Labels and Significance Tests

0 to 6 -3 to 3
Significance Tests
Success 4.86
5.13
F(1,1061)=13.04
p<.001
Moodiness 3.54
3.79
F(1,1064)=7.13
p<.01
Nervousness 3.25
3.43
F(1,1061)=3.87
P<.05
Optimism
5.06 5.15 F(1,1065)=1 ns

Figure 1. Mean Ratings of Success by Numerical Labels and Need for Cognition
5.5
5.4
5.3
5.2
5.1
5
4.9
4.8
4.7
4.6
LNC
4.5
HNC
4.4
0 to 6
-3 to 3









Figure 2. The Percentage of Correct Recall by Numerical Labels and Need for Cognition
70%
60%
50%
40%
30%
20%
0 to 6
10%
-3 to 3
0%
LNC
HNC


4268

---

AAPOR - ASA Section on Survey Research Methods

Figure 3. Attention and Usefulness Rating by Numerical Labels and Need for Cognition
4.8
4.6
4.4
4.2
4
3.8
Attention
Us efulnes s
3.6
LNC, 0 to 6
LNC, -3 to 3
HNC, 0 to 6
HNC, -3 to 3

Figure 4. Percentage of Respondents Inferring “Presence of Failure” by Scale Numerical Values and Need for
Cognition
74%
72%
70%
68%
66%
64%
62%
60%
0 to 6
58%
-3 to 3
56%
LNC
HNC


4269

---