Vous êtes sur la page 1sur 12

Obtaining Purchase Predictions Via Telephone

Interviews
Dean Hini, Don Esslemont and Mike Brennan
Source: Int. Journal of Market Research, Vol. 37, No. 3, 1995

The purpose of this study was to evaluate two methods of obtaining purchase probability data via
telephone interviews; a pre-posted Juster Scale and a verbal 11 category 0-10 scale (Verbal
Probability Scale). The results suggest that both the pre-posted Juster Scale and the Verbal Probability
Scale are viable methods for obtaining purchase probability data via telephone interviews. Both
methods achieved similar response rates and refusal rates, produced results with similar overall levels
of accuracy and appear to pose few problems for either respondents or interviewers. Given the
similarity of the results achieved by the two methods, the preferred option for a telephone survey would
be to use the Verbal Probability Scale since this method is cheaper and can be used with random digit
dialling.

Mike Brennan, Don Esslemont


and
Dean Hini
Massey University, Palmerston North, New Zealand

INTRODUCTION
A wide range of different scales has been developed to enable researchers to estimate demand for consumer
products (see Day et al. 1991, for a review). The evidence to date suggests that the most effective method is
the eleven-point purchase probability scale developed by Juster, shown in Figure 1 (Day et al. 1991; Juster
1966). However, almost all of these scales have been used in a face-to-face situation, and occasionally in mail
surveys (Day et al. 1991), where the scale can be presented to respondents.

Unfortunately, it is not possible to use visual material in telephone surveys, unless it can be delivered prior to the
survey. If not, because of peoples limited ability to remember, response categories in telephone surveys are
often limited to no more than four or five (Aaker & Day 1990; Bradburn & Sudman 1979; Dillman 1978). But
telephone surveys have become the main method of data collection for large samples (Aaker & Day 1990),
particularly among commercial research houses, so there is an urgent need for a reliable method for obtaining
purchase probability data via telephone interviews.

There are several options available if a researcher wishes to use a scale in a telephone survey. One option is to
mail the scale out prior to the survey. However, this option precludes the use of random digit dialling, and
extends the fieldwork by the several days required to compile a mailing list and deliver the mail. An alternative
procedure is simply to ask respondents to choose an answer from an imaginary multiple category numerical
scale, such as from zero to 5 or from zero to 10 (Dillman 1978; Sudman & Bradburn 1982). A third alternative is
to use a branched verbal scale that has two or more levels, such as between zero and 5 or between 5 and 10
chances in 10; between 5 and 8 or between 8 and 10 chances in 10; and so on (Sudman & Bradburn 1982).

Perhaps surprisingly, very few studies have examined the use of these different procedures in telephone
surveys. In one of the few studies that have, Loken et al . (1987) examined the use of an 11-category 0-10 scale
in a telephone survey and concluded that this scale may be a reliable and effective alternative to response
scales with fewer categories. A scale of this form may therefore offer a practical alternative to the Juster scale
for telephone surveys.

More recently, Brennan, Hini & Esslemont (1994) compared three methods for obtaining purchase probability
data via telephone surveys; an 11 category 0-10 verbal scale (Verbal Probability Scale), a branched 11 category
verbal scale, and a pre-posted Juster scale. They abandoned the branched scale, which was found to be too
confusing and laborious for both the interviewer and respondents, although the instructions were quite wordy
and perhaps could be refined. They concluded that both the pre-posted Juster scale and the verbal probability
scale were easy to use and gave comparable results, although the verbal probability scale consistently
produced the more accurate predictions. Furthermore, the response rate to the mailed Juster scale was much
lower (46% vs 74%) and the refusal rate much higher (36% vs 16%) than for the verbal probability scale.
However, the authors stress that the study only involved a small samples, and called for further investigation.

The purpose of this present study was to extend the preliminary work of Brennan et al. (1994), by examining the
effectiveness of a verbal 11 category 0-10 scale (verbal probability scale) and a pre-posted Juster scale for
obtaining short-term purchase probabilities for a range of consumer goods and services, using a larger sample.
The study also examined the effect on response rates of the form of address used on the letters sent with the
pre-posted Juster scale. The paper reports the effects of the form of address, compares both the response rates
and refusal rates achieved by the different modes of delivery of the two forms of scale (pre-posted Juster or
verbal 11-point scale), and examines the predictive accuracy of the two forms of purchase probability scale.

Method
A sample of approximately 350 males and 350 females was randomly selected from the Palmerston North
Electoral Roll. These were cross-checked in the Palmerston North telephone directory, and those names from
the roll that did not match a name or address in the telephone directory were replaced. The final sample
comprised 342 males and 358 females. The electoral roll provided the respondents sex and occupation, which
enabled the assignment of an Elley-Irving socio-economic classification (Elley & Irving 1981). The Elley-Irving
index is a classification of occupations based on median income and educational levels reported in the census.
The scale was devised for educational research purposes to provide a simple measure of parents' socio-
economic status using information their children would know. The index was used in this study to ensure the
different experimental groups were reasonably balanced, to minimise the possibility of any effects being due to
socioeconomic differences.

All names were entered into a database, which was sorted by sex, then Elley-Irving classification. Alternate
names were then allocated to one of two groups, labelled the Mail Group and the No-mail Group. These two
groups were exposed to different methods of obtaining purchase probabilities. In addition, respondents assigned
to the Mail Group were in turn assigned to one of three treatments labelled treatment 1, 2 and 3. These
treatments represented three variations of a covering letter sent to respondents in the Mail Group. The sizes of
the final subsamples are summarised in Table 1.

Procedure
The initial interviews were conducted by six trained interviewers. Allocation of respondents to interviewers was
balanced, as far as possible, across the two groups and three subgroups, and in terms of the sex and Elley-
Irving classification of respondents within these groups. Thus any differences between treatments should not be
due to differences between interviewers, or due to the sex or socioeconomic characteristics of the respondents.

Exactly four weeks after the first interview, respondents were re-contacted and asked whether or not any of the
nine items had been purchased during the preceding four weeks. The study took place between November 13
and December 11, 1993.

Mail group

Respondents were mailed a one-page letter which had the Juster scale printed on the back. They were asked to
keep the letter by the telephone so that it would be ready when an interviewer rang later that week.

Three forms of the letter were tested. T1 had no honorific, T2 was addressed to Dear Householder, and T3 was
addressed to Dear <Mr (or Ms) Surname>. All envelopes were addressed to a specific person <Mr or Ms
Firstname Surname>.

Two or three days after the letters were posted, interviewers telephoned the respondents, identified themselves
by name, and indicated that they were researchers from the Marketing Department at Massey University. Having
contacted the person they required, they explained that they were doing a study on grocery buying and
shopping habits and that this would only take 5 to 10 minutes, and sought the respondents permission to
undertake the interview.

After being read a statement of confidentiality, respondents were asked if they had the letter in front of them. If
not, an arrangement was made to send a replacement letter, and the interview was abandoned. Otherwise, the
interviewer preceded to obtain the purchase probability data.

The specific instructions, read out to respondents, were as follows:

We would like to know what the chances are of you buying certain products during the next four
weeks.

The answers you may give are provided on the Juster scale that is printed on the back of the letter
we sent you. The answers are arranged on a scale a bit like a thermometer. If you are certain, or
practically certain, that you will purchase a product then you would choose the answer 10. If you
think there is no chance, or almost no chance of purchasing, the best answer would be zero. If you
are uncertain about the chances, choose another answer as close to zero or 10 as you think it should
be.

The interviewer then read out the specific questions:

Taking everything into account, what are the chances that you, or anyone else in your household,
will buy the following products within the next four weeks, that is, between now and two weeks
before Christmas. Please give me an answer between zero and 10 from the Juster scale on the sheet
we sent you.

one or more containers of margarine


one or more tins or packets of spaghetti
one or more whole pre-cooked chickens

Now, thinking just about yourself, and taking everything into account, what are the chances that you
personally will do any of the following within the next four weeks. Again, please give me a number
between zero and 10 from the Juster scale on the sheet we sent you.

buy a CD
buy a paperback book (not a magazine)
buy a pair of shoes (for yourself or others)
eat a meal at a restaurant (not takeaways)
go to the movies
travel in a taxi

These particular products and services were used because they had been employed in previous studies (e.g.
Day et al. 1991; Hamilton-Gibbs et al. 1992), had purchase cycles that suited the time frame of the study, and
represented a range of high and low involvement goods and services.

No-mail groups

Interviewers telephoned respondents and introduced themselves and the study in the same way as for the Mail
Group, except that no mention was made of the Juster scale. After obtaining agreement to be interviewed, the
following statement was read out:

We would like to know what the chances are of you purchasing certain products during the next four
weeks.

I would like you to answer on a scale of zero to 10. If you are certain, or practically certain that you
will purchase a product, you would choose the answer 10. If you think there is no chance, or almost
no chance of purchasing, the best answer would be zero.

If you are uncertain about the chances, choose another answer as close to zero or 10 as you think it
should be. You can think of the numbers as chances out of 10. For example, 3 would mean 3
chances in 10 that you would buy the product, while a 7 would mean that there are 7 chances in 10
that you would buy the product, and so on.
The interviewer then read out the specific questions. These were identical to those used with Mail Group,
except that the phrase ...from the Juster scale on the sheet we sent you was omitted.

Purchase rate calculations

Respondents were asked to indicate the chances of purchasing an item during a defined time period by
specifying a number between 0 and 10 from either the Juster scale or the verbal probability scale. Both of these
scales correspond to estimates of the probability of purchase, ranging from 0 to 1. The mean of these individual
probabilities is the predicted proportion of the sample who will buy the item during this period. This is the
predicted purchase rate. While this procedure makes certain assumptions about the linear scaling that might not
hold true (see de Chernatony & Knox 1990), it is the procedure used by Juster (1966) and subsequent studies
(e.g. Brennan et al. 1994; Day et al. 1991). We acknowledge, however, the need to re-examine these
assumptions when sufficient data have been accumulated.

The measure of the actual purchase rate is the proportion of the sample who reported, during the follow-up
telephone survey, having actually purchased the item during the specified time period.

While relying on recall as a surrogate measure of actual purchase, we acknowledge that there is considerable
evidence that recall data obtained from single interviews typically provide over-estimates of actual purchases
(Sudman & Bradburn 1974; Neter & Waksberg 1964; Parfitt 1974). However, there is also evidence that this
telescoping, as it is called, is effectively reduced by using a bounded recall procedure (Neter & Waksberg 1964),
where the beginning of the recall period is bounded by a previous interview. Although our previous interview
gathered purchase probability data rather than recall data, we would expect the effects to be similar, although
this remains to be confirmed. The alternative of using diary panels was not feasible, because of time and cost
constraints. Diary panels were used in an earlier study in this series (Day et al. 1991) which reported predictive
accuracy similar to that obtained in the studies using recall data.

Results

Response rates

The response rates for the two groups are summarised in Table 1. The notable feature is that both the response
rates and the refusal rates for the two groups were almost identical. Both methods achieved response rates of
around 75%, and refusal rates of around 15%. These rates are very similar to those obtained previously by
Brennan et al. (1994) for the verbal probability scale, and are a substantial improvement over those obtained for
the mailed Juster scale. These results indicate that either method would be acceptable as far as response rates
are concerned.

The type of honorific on the covering letter sent to respondents in the Mail Group made virtually no difference to
the response rate and no significant difference to the refusal rate (see Table 1). This is perhaps not surprising,
since the envelope was addressed to a specific person. As the absence of a honorific produced only a
marginally lower response rate than the Dear Mr (or Ms) Surname, this would be the preferred option. It is by far
the most cost-effective option because it does not require mail-merging each letter.
TABLE 1:
RESPONSE RATES AND REFUSAL RATES
Mail group No-
mail
group
T1 T2 T3 Total
(n) (n) (n) (n) (n)

Valid 77 75 75 22 238
Refused 14 19 12 45 49
GNA 9 13 14 36 32
Ineligible 5 - 1 6 1
No contact 15 9 11 35 30
Lost letter - - 1 1 -
Total 120 116 114 350 350
Response rate 72.6 72.8 75.8 73.7 75.1
Refusal rate 13.2 18.5 14.1 14.6 15.5

Note: T1 = no honorific on
letter
T2 = 'Dear
Householder'
T3 = 'Dear <Mr (or Mrs)
Surname>

Predictive accuracy

The accuracy of the predictions obtained by the two methods is summarised in Table 2. The errors in prediction
for the nine items ranged from 0% to 108% , but were mostly less than about 20%.

Table 2 Accuracy of predictions


Product Predicted Actual Difference Error*
(%) (%) (%)
Pre-posted Juster scale

Margarine 68.9 71.8 -2.9 -4


Spaghetti 54.7 62.4 -7.7 -12
Meal 55.3 56.3 -1.0 -2
Shoes 30.6 33.3 -2.7 -8
Chicken 31.6 26.8 4.8 18
Book 21.6 18.9 2.7 14
Taxi 14.0 16.9 -2.9 -17
Movies 34.1 16.4 17.7 108
CD 17.6 11.3 6.3 56
MAD 5.4
MAD (excluding movies) 3.9

Verbal probability scale

Margarine 61.8 71.8 -10.0 -14


Spaghetti 53.9 64.6 -10.7 -17
Meal 43.9 55.2 -11.3 -21
Shoes 32.4 30.9 1.5 5
Chicken 29.1 29.2 -0.1 0
Book 16.5 20.2 -3.7 -18
Taxi 14.1 18.8 -4.7 -25
Movies 27.1 20.2 6.9 34
CD 12.0 10.3 1.7 17
MAD 5.6
MAD (excluding movies) 5.5

Notes 1. Error = (predicted - actual /actual)* 100


2. MAD = mean absolute deviation of the errors

Although the accuracy for both methods varies considerably across the nine products, the mean absolute
deviation (MAD) scores indicate that the mailed Juster scale produced marginally more accurate estimates that
the verbal probability scale, particularly if the results for movies are ignored.

The large error for movies obtained for the Juster scale group (108%) appears to be an anomaly due to
sampling error. There is no other obvious explanation of why this large error should have occurred with one
method and not the other, since the groups were balanced and the instructions identical. Having said this, it is
perhaps unreasonable to expect consistently accurate predictions of demand for products or services which are
purchased or used on an irregular basis. This is especially the case when the decision to use a product or
service, such as going to the movies, is impulsive and depends on a range of situational factors, such as the
weather, the activities of friends, programmes shown on TV, the choice of movies available, and so on.
It is also possible that, with some products and services, respondents were simply reporting their memory of past
behaviour when estimating their future behaviour. This of course is a possibility in any situation where people
are asked about their future actions, and in many cases, basing predictions on past behaviour is a sensible thing
to do. However, one would expect most people to modify their projections in light of what they expect the future
circumstances to be. The implications of this are that predictions are likely to be less accurate when there is less
certainty about the future, particularly for products or services, such as movie visits or taxi rides, that are
typically irregular events undertaken with little forward planning.

This study employed a mix of regularly and irregularly purchased high and low involvement products and
services. This leads to some difficulty when attempting to interpret and generalise from the findings. However,
there are so many ways of categorising products and services, and severe limitations on the number of items
that can be presented to a respondent in a particular study, that the choice of which products or services to use
is somewhat arbitrary. However, our hope (and belief) is that some of the more important findings will become
apparent as information is accumulated over a series of similar studies.

It must be remembered that the use of these techniques is not to forecast future demand (that might be done
better by time-series methods) as to predict future purchases by the population sampled.

Conclusions
The results of this study suggest that both the mailed Juster scale and the verbal probability scale are viable
methods for obtaining purchase probability data via telephone interviews. Both methods achieved similar
response rates and refusal rates, produced results with similar overall levels of accuracy, and appear to pose
few problems for either respondents or interviewers. The response rates of around 75% compare favourably
with those achieved in well conducted mail surveys (Brennan 1992a,b; Brennan, Hoek & Astridge 1991; Dillman
1978), and are rather higher than those typically achieved in face-to-face surveys.

While the accuracy of predictions varied considerably, they are comparable to those achieved in other Juster
scale studies and available evidence would suggest that they are likely to be more accurate than estimates
obtained by the kinds of purchase intention scales commonly used (Day et al. 1991; Juster 1966). Even so,
further research is required to define more fully the situations in which accurate predictions of either purchase
rate or purchase level can be made. There is some evidence that the accuracy of predictions is greatly affected
by whether the product is a regular or irregular purchase, whether the product has a relatively high or low
market share, and whether the respondent is the main purchaser of the product.

The effects of each of these factors requires further investigation, and further efforts to improve the accuracy of
these methods is to be encouraged. The ability to use these methods for telephone surveys should facilitate this
process, since this is a far more cost effective way of increasing sample sizes, which in turn will increase
precision and reduce the errors of prediction.

Given the similar results achieved by the two methods, both in this study and that of Brennan et al. (1994), the
preferred option for a telephone survey would be to use the verbal probability scale. This method is clearly far
cheaper to use than the mailed Juster scale because it does not require the mailout component, and it can be
used with random digit dialling, thereby eliminating the time-consuming task of extracting names from the
electoral rolls and telephone directories.

The purpose of this study was to evaluate the feasibility of two methods for obtaining purchase probabilities via
telephone surveys, and so intentionally had a rather narrow focus. Having demonstrated, across a range of
different products and services, that either method could be used in a telephone survey, the way is open to use
the scales in telephone surveys for other applications already examined using face-to-face interviews. For
example, for many products and services, it is not enough to predict the proportion of consumers who will buy. It
is necessary to predict the mean purchase level, that is, the average number of units bought. Methods for
estimating purchase levels have already been developed (Brennan, Esslemont & U 1992; Hamilton-Gibbs,
Esslemont & McGuinness 1992), and could be employed in telephone surveys using the verbal probability scale.

There is also the possibility of using the methods to estimate demand when this cannot be done using existing
sales data or information about past behaviour. Clearly there is no point in developing a scale for estimating
demand for products such as fast-moving consumer goods, for which reliable sales data are available. Products
from this category (e.g., margarine and spaghetti) were only used in this study for convenience. The real
purpose for developing the scale is to be able to estimate purchase rates for new or hypothetical products for
which no data exist.

A particular application would be to estimate demand at different price points, in order to develop demand curves
that can be used to aid pricing decisions (Brennan, Esslemont & U 1992). Another application would be to
estimate demand associated with alternative advertising or promotional mixes, at the pre-testing stage, to
determine which alternative would most effectively increase sales. Clearly more work needs to be done before
the purchase probability methods could be used with confidence, but evidence from the studies to date suggest
that these methods may well be as reliable as those routinely employed without question at the moment.

In conclusion, the results of this study suggest that both the mailed Juster scale and the verbal probability scale
are viable methods for obtaining purchase probability data via telephone interviews. Given the potential of these
methods for estimating purchase levels as well as purchase rates for new and hypothetical products, for
estimating demand curves, and for pre-testing marketing mix options, the results provide encouragement to
pursue further research into the situations in which the methods can be reliably employed.

REFERENCES
Aaker, D.A. & Day, G.S. (1990) Marketing research, 4th ed. Singapore: John Wiley & Sons.

BradburnN, N. & Sudman, S. (1979) Improving interview method and questionnaire design. San Francisco, CA:
Jossey-Boss Publishers.

Brennan, M. (1992a) Techniques for improving mail survey response rates. Marketing Bulletin, 3, 24-37.

Brennan, M. (1992b) The effect of a monetary incentive on mail survey response rates: new data. Journal of the
Market Research Society, 34, 173-7.

Brennan, M., Hini, D. & Esslemont, D. (1994) Obtaining purchase probability data via telephone surveys: a
preliminary test of two techniques. Marketing Bulletin, 5, 64-70.

Brennan, M. & Esslemont, D.& U, C. (1992) A comparison of two methods for predicting purchase rates and
purchase levels using the Juster Scale. Paper presented at the ANZAM Conference, Western Sydney,
December 6-10, 1992.

Brennan, M., Hoek, J. & Astridge, C. (1991) The effects of a monetary incentive on the response rate and cost-
effectiveness of a mail survey. Journal of the Market Research Society, 33, 229-41.

Day, D., Gan, B., Gendall, P. & Esslemont, D. (1991) Predicting purchase behaviour. Marketing Bulletin, 2, pp
18-30.

De Chernatony, L. & Knox, S. (1990) How an appreciation of consumer behaviour can help in product testing.
Journal of the Market Research Society, 32, 329-47.

Dillman, D.A. (1978) Mail and telephone surveys: the Total Design Method. New York:. Wiley.

Eelley, W.B. & Irving, J.C. (1981) The Elley-Irving socio-economic index: 1981 census revision. New Zealand
Journal of Educational Studies, 20, 115-28.

Hamilton-Gibbs, D., Esslemont, D. & McGuinness, D. (1992) Predicting demand for frequently purchased items.
Marketing Bulletin, 3, 18-23.

Juster, F.T. (1966) Consumer buying intentions and purchase probability. Occasional Paper 99, National
Bureau of Economic Research, Colombia University Press.

Loken, B., Pirie, P., Virnig, K.A., Hinkle, R.L. & Salmon, C.T. (1987) The use of 0-10 scales in telephone
surveys. Journal of the Market Research Society, 29, 353-62.

Neter, J. & Waksberg, J. (1964) A study of response errors in expenditure data from household interviews.
Journal of the American Statistical Association, 59, 18-55.

Parfitt, J.H. (1974) A comparison of purchase recall with diary panel methods, Journal of Advertising Research,
7, 16-31.

Sudman, S. & Bradburn, N.M. (1974) Response effects in surveys: a review and synthesis. Chicago: Aldine.

Sudman, S. & Bradburn, N.M. (1982) Asking questions: a practical guide to questionnaire design. San
Francisco, CA: Macmillan.<

NOTES & EXHIBITS

FIGURE 1:
THE JUSTER SCALE
© Copyright WARC 1995
WARC Ltd.
Americas: 2233 Wisconsin Ave NW, Suite 535, Washington, DC 20007, United States - Tel: +1 202 778 0680
APAC: 20A Teck Lim Road, 088391, Singapore - Tel: +65 3157 6200
EMEA: 85 Newman Street, London, United Kingdom, W1T 3EU - Tel: +44 (0)20 7467 8100

www.warc.com

All rights reserved including database rights. This electronic file is for the personal use of authorised users based at the subscribing
company's office location. It may not be reproduced, posted on intranets, extranets or the internet, e-mailed, archived or shared electronically
either within the purchaser's organisation or externally without express written permission from Warc.

Vous aimerez peut-être aussi