University of Aarhus

1. Introduction

Measurements of political attitudes aimed at predicting voting behavior and other types of political behavior are to a wide extent based exclusively on measurements of attitude direction. It is assumed that such behavior predictions can be made with greater accuracy if, besides the attitude direction, other factors are taken into consideration. The aim of this article is 1) to discuss confidence in behavior predictions at different combinations of direction, intensity, stability, and saliency, and 2) to discuss the concept of saliency and in greater detail operationalizations of saliency.

A prerequisite for prediction of behavior solely on the basis of attitude direction must be that in a given situation a person will show behavior consistent with the attitude direction, e.g. in a given situation a person with a negative attitude towards an object will show hostility or will try to avoid the object.

The social-psychological literature, however, mentions that an attitude, i.e. an individual's predisposition to evaluate an object in a certain way,¹ has several dimensions beyond direction, and we suggest that some of these must play a decisive role in determining the degree of consistency between actual behavior and attitude direction.

However, the number of attitude dimensions is great, and it is generally impracticable in most surveys to consider more than a few dimensions. Such a restriction can of course only be allowed if it is possible from the consideration of a few dimensions to predict behavior with a sufficient degree of accuracy.

2. Concepts

The dimensions assumed to be of importance for the confidence of behavior prediction are intensity and stability. By the intensity of an attitude we mean the degree of emotional engagement, whereas stability is the question of the constancy of the attitude over a long period. Generally, it must be anticipated that the accuracy of the behavior prediction will increase with increasing intensity

and stability. This means, however, that at low values of attitude-dimensions, such as intensity and stability, it will only be possible to make a prediction of behavior if other factors are taken into consideration. In the literature several characteristics of attitude objects are discussed, e.g. saliency, differentiation or inclusiveness, and time-perspective.2 It cannot be precluded that the differentiation of an attitude object in certain cases will be relevant for prediction of behavior. We assume, however, a negative correlation between high degree of differentiation and (1) extremity of attitude-direction and (2) high intensity. As will be seen later the combination of low intensity and non-extreme attitude-direction is assumed to be followed by a low accuracy of prediction of behavior. Furthermore we assume that a high degree of differentiation is positively correlated with high saliency.

The concept of time-perspective of the attitude object is considered as being too vaguely defined and its role in prediction of behavior seems rather unclear. Consequently, in this article we will concentrate on saliency, i. e. the position of an attitude object in an individual's consciousness.

There are other definitions of saliency, however, and the definitions can roughly be divided into three groups. The first group considers saliency as an objectcharacteristicon, whereas the second group uses saliency as an attitude dimension.³ Krech & Grutchfield's definition of saliency as an indicator of importance on a par with intensity and self-involvement can especially be underlined. Thirdly there are several other special applications of the term, e.g. one can talk about a group's, or especially a reference-group's, saliency for an individual.4

One of the most important variations within the first group is T. Newcombe, R. H. Turner & P. E. Converse's definition according to which saliency is a shortterm phenomenon, whereas generalized or permanent saliency is called centrality.5 Centrality in this sense is included in our definition of saliency.

The two attitude dimensions and the object-characteristicon mentioned above
are assumed to be of decisive importance for the accuracy of behavior predictions
based on attitude measurements.

3. Combinations of Variables

If it is assumed that the four variables have the values high and low only, eight combinations will appear as shown in Table I. Such a table should not induce the idea that the different combinations will appear equally often. Moreover, the question of whether some of the combinations are theoretically possible may be raised.

A more important question concerns the occurrence of combinations involving high intensity and low saliency: we doubt the occurrence of attitudes with a strongly emotional engagement whilst the attitude object has at the same time a low position in the consciousness of the individual (combinations 2 and 4).

Concerning the remaining six combinations, Table I shows an evaluation of the
confidence in behavior predictions.

It can be seen from the table that the highest prediction confidence is obtained
at combination 1, and the lowest at combination 8.

A high degree of confidence in behavior prediction must be assumed at combination 3, if the time interval between measurement and behavior is short. In the only other case where it is possible to say something about the prediction confidence, i.e. combination 6, one will unfortunately note that it is low.

The question marks in combinations 5 and 7 indicate that with a combination of low intensity and high saliency it is very difficult to say anything about the behavior prediction. It must be assumed, however, that the behavior prediction has a higher degree of confidence in combination 5 than in combination 7.

Without knowing the frequency of the various combinations it is impossible to estimate the confidence in the behavior predictions made solely on the basis of knowledge of the attitude direction. At any rate it seems rather bold to expect only combination 1 and possibly combination 3 to occur.

4. The Time-Perspektive

As we are primarily interested in predictions of behavior based on opinion polls there will often be a relatively long period of time from measurement to predicted behavior, and this has to be taken into consideration. This effect of time is marked in combination 3, where in the short run the behavior prediction will probably be of great accuracy, but will decrease over time. We have tried to illustrate this in Figure 1.

Further, it appears from Figure 1 that we assume that, with combination 1, the
prediction confidence within a not unreasonably long time will be constant and
high.

For the combinations 6 and 8 we also assume a constant but low prediction
accuracy.

A special problem of behavior predictions based on measurements of attitude direction, intensity, stability, and saliency by survey techniques is that these measurements often refer to one point in time and give no information about the value of these variables over time. Predictions on the basis of variables of which one knows only one value are based on the assumption that the value of the variable is constant over time. In cases where the interval between measurement and predicted behavior is short, a breach of the mentioned condition will be much less important than it will in long-term predictions.

5. Variation along the Direction Continuum

When commenting on Table I and Figure 1 we have presupposed a given position on the positive-negative continuum. In Figure 2 the relation between the confidence of behavior predictions and attitude direction is shown by different combinations of intensity, stability, and saliency. From Figure 2 it appears that even at combination 1 one will have to expect a low confidence in behavior prediction if the individual is placed near the neutral point on the positive-negative continuum.

Further., as shown in the Figure, at combination 8 one will generally have a low confidence in behavior predictions. For both curves it may be said, however, that parts of them may not exist at all or very rarely. This applies to the curve of combination 1 which hardly exists at the neutral point and just around this point, and to the curve for combination 8 which hardly exists for extreme direction values. In both cases these modifications can be supported by reference to the often ascertained U-shaped relation between intensity and direction.

6. Operationalizations

The above discussion should make it clear that intensity, stability, and saliency are
very important factors for determining confidence in behavior prediction.

While there is a tradition for the measurement of intensity, this is not the case
for stability and saliency.

Intensity is often operationalized by questions like: "How deeply engaged are you in this?" If "emotionally" was added to this question it would have high internal validity as well as a satisfactory reliability. The only possibility for ascertaining stability is repeated measurement, and in many surveys this is impossible for practical reasons. However, it cannot be excluded that in certain cases one might obtain a measure of the stability of the attitude by asking questions concerning the constancy of the attitude.

Operationalization of saliency is an attempt to find methods of measurement to give us information about the position of an attitude object in an individual's consciousness. As there is no standard way of measuring saliency, we will discuss a number of different approaches.

1. Level of Information: Saliency measured by obtaining an individual's information
about the object.

a) internal validity: even if it must be assumed that a high positive correlation
between high information about an object and the saliency of this object, and
similarly with low information and low saliency, can often be ascertained, one
might find examples, however, where this correlation does not exist.
b) reliability: no particular problems.

2. Interest: Saliency measured by obtaining an individual's interest in the object.

a) internal validity: interest in an object will, assuming that prestige distortion can be kept under control, presumably be a valid measure of saliency, but it will be difficult to keep prestige elements out of the measurement. Perhaps more important, validity will be low in certain cases where the relation to the object is involuntary.

b) reliability: no particular problems.

3. Exposure: Saliency measured by the degree to which an individual is exposed
to communication about the object.

a) internal validity: a high degree of exposure may be an expression of high saliency, but this need not be the case. That a person is exposed to a steady flow of communication about an object does not guarantee that the messages are received, and even if this is the case it does not necessarily mean that the object has a high position in the individual's mind.

b) reliability: no particular problems.

4. Awareness: Saliency measured by the degree to which the attention of the individual
is drawn to the object. This method will not be discussed, as it will be very
difficult to separate it from the three previous methods.

M. K. Jennings & H. Zeigler6 mention a method which is difficult to place in one of the types mentioned, because both the terms "follow what's going on" and "interested" are used: "Some people seem to follow what's going on in government and public affairs most of the time, whether there's an election going on or not. Others aren't that interested. Would you say that you follow what's going on in government and public affairs most of the time, some of the time, only now and then, or hardly at all?" Those selecting one of the first three alternatives are considered to be members of the attentive public. After handing the respondent a card showing the four levels of public affairs, he was asked to rank them in this fashion: "Which one of these kinds of public affairs do you follow most closely?" "Which one do you follow most closely?" "Which one do you follow least closely?" With first, second, and fourth ranks thus determined the residual level automatically occupied the third rank. As mentioned under the previous three methods the internal validity of this method, too, may be questioned.

5. Participation: Saliency measured by the degree to which an individual is behaviorally
involved with the object.

a) internal validity: though it might be reasonable to assume that when an individual is behaviorally involved with an object, this object will also have a high position in his consciousness, one cannot conclude to the contrary that low behavioral involvement with an object means that the object has a low saliency. b) reliability: no particular problems.

6. Number of "Don't knows": Saliency measured by how often an individual has
answered "don't know" to questions concerning the object.

a) internal validity: this measure of saliency must be supposed to be applicable
in many cases, though often with the risk that a "don't know" answer might
be an expression of high saliency combined with, for instance, high information.
Moreover, the method has the weakness that in many cases the respondent will
feel pressed to answer a question about an object which does not have a high
saliency, i.e. cases where the respondent "ought to" answer "don't know".
b) reliability: no particular problems.

7. "No opinion": Saliency measured by giving the respondent the possibility of
saying that he is unable to answer questions about the object in question or to take
a position on statements concerning the object.

a) internal validity: the weakness of this method is that high saliency combined with high information and interest might even be the reason why the respondent feels unable to take a position, just as the missing opinion might also be due to low saliency. It cannot be excluded either that prestige distortions might occur. b) reliability: reliability may be low, as it may be difficult to ensure that two interviewers give the respondent the same possibility of declaring himself unable to take a position.

This way of operationalizing saliency is used by Philip E. Converse⁷: "At the beginning of the issue item the interviewer said, 'Of course, different things are important to different people, so we don't expect everyone to have an opinion about all these. If you don't have an opinion, just tell me that'. The initial question after each item was introduced, before the opinion was actually asked, was 'Do you have an opinion on this or not?'".

Besides the mentioned weaknesses of the method, it can further be underlined
that importance need not coincide with saliency (see method no. 8).

8. Importance: Saliency measured by the respondent's answer to questions about
importance.

a) internal validity: questions concerning problems such as: "What is the most important problem facing this nation?" might have a low internal validity. A number of issues can be very salient to the respondent, but only two or three issues are mentioned. A low internal validity will also occur in cases where the

respondent is pressed to mention an issue with low saliency.8 Generally one
cannot expect saliency and importance to be coincident. Issues which are perceived
as being unimportant might be very salient to people.

b) reliability: no particular problems.

9. Spontaneous reference: Saliency measured by the respondent's inclination to
bring up a subject spontaneously during an interview.

a) internal validity: apart from situations where the respondent's attitudes are
considered socially unacceptable by himself or the like, this method can be
assumed to have a high internal validity.

b) reliability: obviously the unstructured interview has a lower reliability than the
structured interview.

For reasons of time and economy it is possible to use this method in very few
cases, just as the coding will cause many difficulties in cases where a tape recorder
is used at the interview to improve reliability.

D. Krech & R. S. Crutchfield9 discuss a survey made in Germany in 1946 aimed at measuring the saliency of the bombings. Here 35 questions were asked about other war-subjects than the bombings. The measure of saliency was the point of time in the interview when the respondent would spontaneously mention the bombings and the frequency with which the respondent would spontaneously refer to the bombings. A similar method was employed in the USA during the war in connection with surveys of people's attitude towards the issued war bonds. Here also the initial questions concerned other topics like inflation, the development of the war, etc. A variant of this method is often used in election surveys. Thus D. Butler & D. Stokeslo take up a subject, and through an open question examine which parts of the object have the highest saliency: "Now I would like to ask you what you think the good and bad points about the political parties are", "Is there anything in particular that you like about the Conservative Party?" "What is that?" "Anything else?" A similar method has been employed in Swedish and Norwegian election surveys.

Besides the weaknesses mentioned above, the method only provides saliency measurements
of those parts of the objects to which the respondent takes the most
extreme position.

Among the 9 above-mentioned operationalizations of saliency, No. 9 appears to be theoretically the most satisfactory. This method can be difficult to use in some survey situations and, moreover, it cannot be excluded that some of the other methods are preferable in connection with special problems. Until sufficient empirical results of saliency measurements are available, we are unable to recommend any one of the 9 methods of measurements. On the contrary it must be considered desirable that a study includes as many different saliency measurements as possible.11 It is especially desirable that experiments should be made in order to illustrate both the applicability of the individual saliency measures and the relations between them

as well as the relation between saliency and the above-mentioned attitude dimen
sions.

Above we have postulated that the saliency of an attitude object is a decisive factor in behavior predictions based on attitude measurements, and we have listed a number of operational definitions of saliency which, however, are assumed not to be satisfactory in many cases.

Primarily, however, this article should be seen as an attempt to bring about a
much needed discussion about the concept of saliency.

---

NOTES

1 Katz & 1959.

Stotland, 2 Smith et al., 1970.

3 Scott, 1968; Newcombe, 1952; Krech & Crutchfield, 1948.

4 Burnstein & Mcßae, 1965.

5 Newcombe et al., 1965.

6 Jennings & Zeigler, 1970.

7 Converse, 1964.

8 Unpublished memo from S. Bessmer to R. A. Brody, both at Stanford University.

9 Krech & Crutchfield, 1948.

10 Butler & Stokes, 1969.

11 E.g. Hansen, Petersen & Redder, 1969. This study includes measurements of interest, information, awareness, and participation. These variables can be used as indicators of saliency as well as to construct a composite measure of saliency.

NOTES

1 Katz & 1959.

Stotland, 2 Smith et al., 1970.

3 Scott, 1968; Newcombe, 1952; Krech & Crutchfield, 1948.

4 Burnstein & Mcßae, 1965.

5 Newcombe et al., 1965.

6 Jennings & Zeigler, 1970.

7 Converse, 1964.

8 Unpublished memo from S. Bessmer to R. A. Brody, both at Stanford University.

9 Krech & Crutchfield, 1948.

10 Butler & Stokes, 1969.

11 E.g. Hansen, Petersen & Redder, 1969. This study includes measurements of interest, information, awareness, and participation. These variables can be used as indicators of saliency as well as to construct a composite measure of saliency.

REFERENCES

Burnstein, E. & Mcßae, A. V.: Some Effects of Shared Treat and Prejudice in Racially
Mixed Groups (in Steiner, I. D. and Fishbein, M.: Current Studies in Social Psychology,
1965).

Butler, D. & Stokes, D.: Political Change in Britain, 1969.

Converse, P. E.: New Dimensions of Meaning for Cross-Section Sample Surveys in Politics
(in International Social Science Journal, Vol. 16, no. 1, 1964).

Hansen, P., Petersen, N. & Redder, K. W.: Foreign Policy Altitudes of the Danish
Population (Pretest), 1969.

Jennings, M. K. & Zeigler, H.: The Salience of American State Politics (in the American
Political Science Review, Vol. LXIV, No. 2, 1970).

Katz, D. & Stotland, E.: A Preliminary Statement to a Theory of Attitude Structure and
Change (in Koch, S. (ed.): Psychology: A Study of a Science, Vol. 3, 1959).

D. &R. S.:and ProblemsSociall94B.

Krech, Crutchfield, Theory of Psychology,
Newcombe, T., Turner, R. H. & Converse, P. E.: Social Psychology, 1965.

Scott, W. A.: Attitude Measurement (in Lindsey, G. and Arenson, E.: The Handbook of

Smith, M. B. et al.: Some Properties of Attitudes (in Kessel, J. H. et al.: Micropolitics,
1970).