Intentions to Buy As Predictors of Buying Behavior

Citation:

Susan P. Douglas and Yoram Wind (1971) ,"Intentions to Buy As Predictors of Buying Behavior", in SV - Proceedings of the Second Annual Conference of the Association for Consumer Research, eds. David M. Gardner, College Park, MD : Association for Consumer Research, Pages: 331-343.

Proceedings of the Second Annual Conference of the Association for Consumer Research, 1971     Pages 331-343

INTENTIONS TO BUY AS PREDICTORS OF BUYING BEHAVIOR

Susan P. Douglas, Temple University

Yoram Wind, University of Pennsylvania

[The research on which this paper was based was funded in part by a Grant in Aid, No. 400-101-64, from Temple University. The authors wish to acknowledge the assistance of Rona Zevin for the computer runs associated with this project.]

The use of attitudes as predictors of purchase behavior has long been a source of some dispute. Several reviews of the literature on attitudes and behavior (Wicker, 1969; McGuire, 1966) have concluded that attitudes are unrelated or only slightly related to overt behavior. Others, however, have contended that certain types of measures of attitudes, and particularly purchase intentions, may effectively be used as predictors of buying behavior.

Fishbein (1971), for example, suggests that although attitudes toward &n object (e.g. attitudes toward a brand), may not be good predictors of specific acts, attitudes toward performing a given behavior with regard to an object (e.g. attitudes toward buying a particular brand in, a given situation) will usually be related to the particular behavior in question. More specifically, modifying Dulany's theory of propositional control, Fishbein (1967), proposed a model for predicting behavior based on expressed behavioral intentions which reflect the attitude toward the act rather than the object.

The utility of measures of behavioral intentions as predictors of behavior has been empirically verified in a study by Ajzen and Fishbein (1970) and in a number of studies of the Survey Research Center of the University of Michigan (Mueller, 1957). Indeed, many firms who use intentions to buy in evaluating new product concepts, claim that these measures provide a useful predictor of actual purchases at the aggregate level.

Although these findings suggest that there is a relationship between purchase intentions and behavior, this link is based on the assumption that carrying out intentions is under the individual's control, and that expressed intentions are related to the individual's subsequent behavior. Furthermore, the strength of the relationship and thus the predictive accuracy of the purchase intentions may vary depending on the a) specificity of the purchase intention, i.e., an intention to purchase a generic product vs. a particular item, b) the novelty of the item, c) the particular measure of purchase intentions, and d) the time between the measure of intentions and behavior. The specificity of the measure may be a critical factor. Fishbein (1971) has postulated that "The more specific the measure of intention is to the behavior that is to be predicted, the higher the intention-behavior correlation will be." In operations terms this implies that the correlation between a general measure and general behavior, i.e., intention to buy a product --- purchase of a product, will be higher than between a specific^ measure and specific behavior, i.e., intentions to buy a specific style of a product --- purchase of this style. The novelty of the item may affect the reliability of the measure. In this context it is hypothesized that the more novel the item, the higher the predictive efficacy of the purchase intentions. Different measures of purchase intentions may also be expected to vary in relative efficiency. Finally, the time interval between the measurement of the purchase intentions and the measurement of behavior may affect the level of correlation between the two measures.

Although various studies have been conducted to test the reliability of purchase intentions as predictors of behavior, little attention has been devoted to comparing the reliability and relative efficiency of different measures of purchase intentions.

The purpose of this paper is, thus, to examine the impact of each of these four factors on the predictive efficacy of purchase intentions, based on a longitudinal study of women's attitudes and behavior toward fashion items.

THE DATA

A study of attitudes and behavior toward fashion items was conducted in the Philadelphia area for an eight week period during the fall of 1970. The sample consisted of 82 women from different socio-economic backgrounds, and with different physical builds and included both married and single, working and non-working women.

At the beginning of the study period a questionnaire was administered by personal interview covering attitudes toward and intentions to buy items of clothing. Respondents also provided data on their socio-economic, demographic and life-style characteristics.

Intentions to buy were assessed using a variety of measures. Respondents were asked to rank order eleven fashion sketches which included coats, dresses, skirts and pant suits in above knee, knee length and midi lengths according to their overall preference, intentions to buy, girl friends' preference, male friend's preference, and suitability for everyday and special occasion.

Respondents were also asked to indicate the number of items in each category of clothing (coats, suits, dresses, skirts, pant suits, etc.) they intended to buy during the fall season, the specific lengths (above knee, knee length, mid calf, and ankle length) and occasion (everyday or special occasion) for each category and the total amount they planned to spend. Purchase intentions, specifically, with regard to midis were measured on a five point intentions to buy scale, and attitudes of both girl friends and male friends toward midis were also examined.

During the eight week study period the panel kept records of all shopping trips relating to fashion clothing items. For each shopping trip the number, length, occasion and price of items tried on and purchased in each category was recorded. Half of the sample also filled in prior to each trip, specific purchase intentions for that trip. These diaries were collected on a bi-weekly basis.

At the end of the study period respondents again rank ordered the eleven fashion sketches on the six preference scales. In addition, intentions to buy a midi-coat, dress or skirt in the forthcoming season were measured using a five point intentions to buy scale.

These data are subject to a number of limitations.

1. The size of the sample is small, from a single geographic area and does not cover the whole range of socioeconomic, demographic and psychographic variables which may be expected to influence purchase behavior.

2. The length of the study period is short, particularly in view of the nature of the item considered - clothing.

3. The diary data probably underestimates the number of shopping trips actually made and possibly the number of items tried on. There was also no control over whether purchase intentions for each trip were in fact filled in prior to the trip or afterwards.

DATA ANALYSIS

The relationship between the set of intentions to buy and the set of actual purchase behavior was assessed using three related analytical procedures: a) cross tabulations, b) regression analysis and c) canonical correlation.

The cross tabulations provided an initial understanding of the nature of the relationships between each of the intentions to buy measures and the various purchase measures. More rigorous analysis of the nature of this relationship was then conducted using multivariate statistical techniques. A regression analysis was undertaken to assess the statistical association between each of eleven measures of midi purchase intentions (the independent variables) and the purchase of a midi (a dummy dependent variable). Separate regressions were run for each of the independent variables as well as a multiple regression with all eleven independent variables.

The eleven measures included: a behavioral measure -- the proportion of midis of the number of items tried on; two projective measures -- attitude of girlfriends and of male friends toward midis; two direct intentions to buy measures -- a five point intentions to buy scale and a four point index based on intentions to buy a midi in each category; and six relative measures -- the average ranking of the four midi fashion sketches on the six preference scales -overall preference, probability of purchase, girlfriends' preference, male friends' preference, everyday suitability and special occasion suitability.

A canonical analysis was then used to examine the relationship between the two sets of purchase intentions and behavior. The predictor and criterion variables included dollar amount, number of items, length, and occasion planned and actually purchased across product categories.

RESULTS

The results of the study confirm the findings of previous research that measures of purchase intentions do provide reasonably good predictors of behavior. There was, however, considerable variation in predictive efficacy among the various measures. In particular, each of the four factors examined, the specificity of the measure, the novelty of the item, the type of measure, and the time interval between measurements appeared to affect the efficiency of the measure.

The Specificity of the Measure

The findings of the comparison of general and specific measures of purchase intentions for the five product categories is shown in Figure 1. This shows that while the predictive efficacy of general measures (i.e. intention to buy a product such as a skirt or coat) is slightly greater than the predictive efficacy of specific measures (intentions to buy a specific product style (such as a midi skirt), it varies with the particular product. For three out of five product lines (pant suits, skirts, dresses) the percentage of those intending to buy who actually bought the product was slightly higher than the percentage of those who intended to buy a specific item (product x for special occasion or everyday use or in an above knee, knee or midi length) who actually bought it. For suits and coats the reverse was generally true.

There is also considerable variation in the predictive efficacy of various specific measures depending on the nature of the predictive scenario.

With the exception of suits, purchase intentions for everyday occasions were more accurate than the ones for special occasions, again probably reflecting a higher degree of selectiveness or uncertainty with regard to less commonly purchased items.

The findings also show a considerable variation in predictive efficacy between the various clothing categories for both general and specific measures. In the case of the general measures, the proportion of those who intended to buy a particular item who actually bought it ranged from 14% for suits to 64$ for dresses. This variation across categories appears to be related to frequency of purchasing, cost of the item and other situational factors such as climate and the trend of the fashion cycle.

FIGURE 1

PROPORTION OF THOSE INTENDING TO BUY VARIOUS FASHION ITEMS WHO ACTUALLY BOUGHT THEM: A COMPARISON OF GENERAL AND SPECIFIC INTENTION MEASURES

The frequently purchase lower unit cost items, such as skirts and dresses, had a high percentage of subjects who followed their intentions. For coats, on the other hand, which are higher cost, less frequently purchased items, only one fifth of the 66% of the sample who intended to buy a coat had actually bought one by the end of the study period.

This may be due in part to uncertainty about fashion lengths, and in part to the Season. Since the last diary was collected at the end of November, those intending to buy a heavy coat for the winter may not yet have purchased it. Alternatively, uncertainty with regard to what styles would be fashionable may have caused some hesitation. Since the purchase of a midicoat is often viewed as a prerequisite to the purchase of a midi skirt or dress, and in addition the purchase of a coat is typically a relatively expensive item, more conservative purchasers might tend to wait and see what lengths were adopted.

The same dependency on product line is also evident in the specific measures of purchase intentions with regard to length and style. In the case of intentions to buy specific lengths, for example, purchase intentions tended to be more accurate for skirts and dresses than for coats and suits, though in all cases these were quite low.

The Novelty of the Item

One factor which may account for some of the variability in intentions to buy across different product lines is the novelty of the item or style.

The results of this study suggest that while purchase intentions for the most commonly purchased length -- above knee -- were more reliable than for the new midi-length or for the in-between knee length, purchase intentions for novel items -- pant suits and midis -- were more accurate than more common items such as dresses and skirts.

Figure 2 compares the statistical association between intentions to buy and purchase behavior for the novel items with the mere common items.

FIGURE 2

THE PREDICTIVE EFFICACY OF INTENTIONS TO BUY "NOVEL" AND "COMMONLY PURCHASED" ITEMS

In the case of the midi, the R2 is substantially higher than for the skirt or the dress. In the case of the pant suit it is only slightly higher. This is somewhat surprising since Figure l showed that a substantially higher proportion of those planning to buy pant suits actually bought them compared with the other categories. This may, however, be a function of ambivalence with regard to lengths rather than the novelty of the item. A purchase of a pant suit avoided the problem of choosing a specific length. It thus provided a compromise solution for those who were uncertain which lengths would be fashionable, avoiding both the risk of having purchased a midi if the fashion did not catch on, or alternatively if the midi did become fashionable, of having an item which would be out of fashion by the following year.

The Type of Measure

One of the key research issues in the use of "intentions to buy" as a predictor of purchase behavior is the appropriate or most effective way of measuring intentions to buy. In this study four different types of measures were used to predict purchase of the midis.

(a) A behavioral measure -- the proportion of midis tried on

(b) Two measures of intentions to buy a midi -- five point intentions to buy scale, and a simple statement of intentions to buy or not buy a midi

(c) Six preference measures -- the average rank of 4 midi sketches compared to the position of other fashion items

(d) A projective attitude measure -- a girlfriend's and male friend's attitude toward the midi

A composite measure using ten of the single measures to predict purchase of a midi was also developed.

The results of the study show substantial differences in the predictive efficacy of the eleven single measures. The results of the simple regression analyses using each of these measures singly as a predictor of purchase or nonpurchase of a midi are shown in the first two columns of Figure 3 and reveal a substantial range in R2. The behavioral measure -- the proportion of midis tried on -- provided the best predictor of purchase behavior with a R2 of .72. The two direct measures, the five point scale and the intentions to buy across categories also provided relatively good predictors of behavior with R2 of .55 and .54 respectively. All of the six relative attitude measures were less efficient with R2 ranging from .25 for girl friends' preference to .41 for probability of purchase. The least efficient predictors were the two projective attitude measures with R2 of .20 each.

This suggests that simple direct measures of purchase intentions can provide relatively reliable indicators of behavior and at least in this context are more effective than relative or projective measures. The relatively poor predictive efficacy of the relative measures -- rank ordered data -- may be due to the fact that respondents typically bought more than one item, and thus a rank order of items from different categories did not necessarily reflect the desired purchase assortment nor the purchase priority. Another possible reason for this relatively poor performance is the averaging procedure used here. Utilizing the complete rank order data may lead to better predictive efficacy.

A somewhat more surprising finding is the low R2 for the two measures of girl and male friend's attitudes. Since acceptance by the peer group is generally considered an important factor in fashion adoption, one might have expected a stronger relationship between perception of friends' preference and the respondent's behavior. As, however, the study was undertaken at the beginning of the midi fashion cycle, those who bought midis might be classified as "innovators" or "early adopters". Given some earlier evidence that the early adopters tend to be fairly independent, they may be less affected by attitudes of their friends toward fashions.

The composite measure using 10 of the predictor variables proved superior to any single measure. The R2 of .83 only represented, however, a slight improvement over the best single variable -- the behavioral measure -- with an R2 of .72.

Comparing the standardized p coefficients for the multiple regression with the R2 for the independent regression runs in Figure 3 shows some differences in relative importance of the various measures, particularly the six relative preference measures. While based on the multiple regression both the behavioral measure and the 5 point scale remained the most effective predictors, the intention to buy a midi in each category decreased in importance. Similarly, the relative importance of the six preference measures tended to shift. In the independent runs, probability of purchase followed overall preference had the highest R2 of the relative measure while in the multiple regression, male preference had the highest 2 coefficient. In both sets of analysis, however, the two projective measures were the least effective measures.

FIGURE 3

RESULTS OF THE REGRESSION ANALYSIS OF PURCHASE/NON PURCHASE OF A MIDI

While the preceding analysis has been concerned with the use of different measures to predict a single dependent variable -- purchase or no purchase of midi -- the canonical correlation analysis provides insight into the association between the set of intention measures and a set (not a single measure) of actual purchase behavior. This enables the researcher to assess whether the two sets (of purchase intentions and behavior) are independent and which variables in each of the two sets contribute most to the between set association.

The correlation matrix for the selected eight intentions and eight purchase measures is shown in Figure 4. In looking at the correlation matrix we note a low degree of intra-set correlation. This is particularly marked among the predictor variables. The only exception being intentions to purchase items for special and for everyday occasions that are correlated with an overall summary measure of intentions to buy any of the five product lines. Only five of the criterion variables show any degree of intercorrelation. The number of items purchased is correlated with two other correlated variables -- purchases of items with above knee length and with items for everyday usage. Similarly, purchase of above knee length items is correlated with total amount spent on clothing during the study period. This suggests, therefore, that composite measures are not likely to be more effective than simple measures.

The overall correlation between the two sets of variables is presented in Figure 5. This shows a fairly high degree of overall correlation between the two sets of variables with a first canonical correlation of .753 (Wilks Lambda of .14). The first (maximally correlated) associated with the first linear compound are shown in Figure 6. With respect to the predictor set we note that the highest coefficient (.861) is associated with intention to buy a midi length item followed by intentions to buy items at above knee length and total amount planned to be spent on clothing. The highest criterion variable is actual purchase of a midi item followed by number of product lines purchased. These findings to a large extent confirm the results of the earlier analysis that purchase intentions for specific items and for novel items have a higher predictive efficacy.

FIGURE 4

CORRELATION MATRIX OF 8 INTENTIONS TO BUY AND 8 PURCHASE MEASURES

FIGURE 5

CANONICAL CORRELATION FOR THE FIRST THREE LINEAR COMPOUNDS

FIGURE 6

STANDARDIZED CANONICAL COEFFICIENTS FOR CRITERIA AND PREDICTOR VARIABLES ASSOCIATED WITH THE FIRST LINEAR COMPOUND

The Time Between the Measure of Intentions and Behavior

In general one may anticipate that the longer the time interval between the measure of intentions and the measure of behavior the lower the correlation between the two will be. Clearly, the longer the time interval the greater the probability that additional factors will change the intentions. However, in the case of clothing items, where several items within a given product category may be purchased during a given time period, the longer time interval, the higher the probability that an item in a particular category will be purchased.

An analysis of the link between purchase intentions and behavior over two different time intervals suggests that this may depend on the specificity of the measure. Figure 7 compares the relation between purchase intentions at the beginning of the study period and behavior throughout the 8 week period with that between purchase intentions prior to a shopping trip, and purchase behavior on that trip.

In the case of general purchase intentions, the percentage of those intending to buy a given product line who actually bought,aggregated across all product items, was slightly higher for the longer than the shorter time period. When broken down by individual categories, pant suits and dresses were substantially higher, for skirts slightly lower, and for suits and coats significantly lower.

In the case of specific purchase intentions, i.e., to buy a specific length or for a particular occasion, the measures for the shorter time period were invariably more effective.

FIGURE 7

PURCHASE INTENTIONS AND BEHAVIOR UNDER SHORT AND LONG TIME INTERVALS

CONCLUSIONS

Despite the limitations of this study, in particular the small sample and relatively short period of observation, a number of conclusions may be drawn:

1. Even in the highly uncertain fashion climate of fall 1970, purchase intentions provided a relatively efficient predictor of behavior. In particular, a simple measure -- a 5 point scale -- had relatively high predictive ability. There was considerable variability in the predictive efficacy for different product categories and in relative efficiency of various measures of purchase intentions.

2. Measures of specific intentions were less accurate than measures of general intentions as predictors of behavior over an extended period of time, but more accurate than general measures as predictors of immediately subsequent behavior (i.e., a specific act)

3. Purchase intentions for novel fashion items were more accurate than purchase intentions for more common items.

4. Behavioral measures, such as trying on items, appear to have significantly better predictive power than verbal measures (i.e., stated purchase intentions or attitudes).

5. As might be expected, composite measures appear to be superior to single measures, though the difference between the best single measure was not substantial. Furthermore, there was little evidence of interrelation between various measures suggesting that more complex measures may not necessarily provide improved results.

In brief, therefore, the results of the study tend to confirm previous findings that behavioral intentions may effectively be used to predict purchase intentions. In addition, they suggest that simple measures, such as the 5 or 7 point scales currently used by various firms, and behavioral measures of intention to buy, may be the most appropriate types of measures. Further research on a larger sample under more controlled conditions is, however, needed before more definite conclusions can be reached.

REFERENCES

Ajzen, I. and Fishbein, M., Attitudinal and normative variables as predictors of specific behaviors: a review of research generated by a theoretical model. Paper presented at the Association for Consumer Research Workshop on Attitude Research and Consumer Behavior, University of Illinois, 1970.

Fishbein, M., Attitude and the prediction of behavior. In M. Fishbein (Ed.) Readings in attitude theory and measurement, New York: John Wiley, 1967.

Fishbein, M., The relationships between attitudes and behaviors. In K.K. Sereno and C.D. Mortensen (Eds.) Advances in communication research, New York: Harper and Row, 1971, in press.

McGuire, W.J., The nature of attitudes and attitude change. In G. Lindzey and E. Aronson (Eds.) The handbook of social psychology, 2nd ed., Vol. 3. Reading, Mass: Addison-Wesley, 1969.

Mueller, E., Effects of consumer attitudes on purchases, The American Economic Review, 1957, 47 946-965.

Wicker, A.W., Attitudes vs. Actions: the relationship of verbal and overt behavioral responses to attitude objects , Journal of Social Issues, 1969, 25, 41-78.

----------------------------------------