Rick, who is unhappy with the increased frequency of repairs that his current car requires, intends to buy a new car to replace it within the next six months. Sandy, on the other hand, has her mind set on buying a bicycle in the coming month to help her move around more freely within the city. Rick and Sandy plan to travel together for a vacation in autumn (three months from now) to visit the lakes in northern Italy. But will they actually follow on their intentions and make the purchases, or the necessary reservations? They may only contemplate their plans to purchase at this time, or they could already feel confident about making them. Situational factors may surface between the time of expressing their purchase intentions and the time for executing them. An intention sets a path to make a product or service purchase, creating the ‘right mindset’ or an expectation, but it does not guarantee that the intention will come true in an actual purchase (or performing an activity).
The gap that often occurs between a purchase intention and actual purchase behaviour of consumers is an issue that has been preoccupying marketing researchers and managers for decades long. It can be disturbing when forecasting demand or sales since the gap is likely in effect to lower the accuracy of predicting consumer purchases in a future period. Different techniques and models have been proposed over the years by academic and practitioner researchers for measuring intentions, while confronting the interference that such a gap may cause to predictions, and how it can best be resolved. Configuring a methodology for predicting product purchases involves several steps, such as choosing the measurement instrument, identifying plausible causes for a gap between intention and behaviour, and devising an adjustment of predictions to reduce the gap effect and increase the accuracy of predictions.
We will focus here on two types of intentions questions: perceived likelihood of purchase and perceived probability of purchase, through the examples given below [1]:
| How likely are you to buy a new car in the next 12 months? | During the next 12 months, what do you think the chances are that you will buy a new car? |
| Will definitely buy | Certain, practically certain (99 in 100) |
| Will probably buy | Almost surely (9 in 10) |
| May or may not buy | Very probably (8 in 10) |
| Will probably not buy | Probable (7 in 10) |
| Will definitely not buy | Good possibility (6 in 10) |
| Fairly good possibility (5 in 10) | |
| Fair possibility (4 in 10) | |
| Some possibility (3 in 10) | |
| Slight possibility (2 in 10) | |
| Very slight possibility (1 in 10) | |
| No chance. almost no chance (1 in 100) |
Examples of two types of intentions questions
The scale of the question on likelihood of purchase is more general, where the interpretation by respondents is based primarily on the wording of ‘probably’ and ‘definitely’. The scale of the question on probability of purchase (proposed by Juster) is more detailed and concrete: it specifies quantitative expressions of probability, yet it also aids respondents with corresponding verbal meanings.
Morwitz [1] recommends the use of a probability-based question to measure intentions, following research which suggests that asking respondents to assess the probability of purchase will be more accurate than other measures of intentions. In particular, although the responses to the two scales above are positively correlated, information from the probability-of-purchase question subsumes the information contained in the likelihood-of-purchase question, but not conversely (i.e., think of a sub-range of response levels of probability chosen by respondents which correspond to a single category of the likelihood question). She notes, however, that studies do not provide a clear answer whether the advantage is due to greater detail of the scale or the reference to probabilities. Researchers do not all agree about the use of probability scales. Counter arguments doubt the ability of consumers to reliably give answers on a ratio scale (as in the case of probability), assess probabilities in particular, or be specific enough in choosing their response on an 11-point scale. It should be noted, however, that the verbal meanings as proposed by Juster can help respondents in comprehending the probability options.
Using intentions data requires additional considerations. Morwitz [1] suggests that researchers should distinguish between two groups of respondents: intenders vs. non-intenders, based on their answers. There can be significant differences in behaviour between these groups. On the one hand, not all intenders (e.g., will definitely buy) can be expected usually to fulfill their intention. On the other hand, some of the non-intenders (e.g., probably and even definitely will not buy) can be eventually found to behave inversely and make a purchase. The latter possibility is more often neglected, which may have the greater impact on the prediction (i.e., because the group of non-intender respondents is usually the larger). For instance, if a researcher does not account for 10% of those not intending to buy a new durable product (e.g., a car, a TV) but who actually do buy the product, the purchase incidence rate may be substantially underestimated.
One of the behavioural adjustment approaches reviewed by Morwitz is using weighted box methods. With respect to likelihood of purchase, different conversion rates may be applied to percents of responses on part (e.g., top two boxes) or all five of the response categories in the example above. However, it is essential to consider differences between types of products (e.g., durables, packaged goods) that can imply the application of different relevant schemes of conversion rates (weights) to the response ‘boxes’. A weighting scheme for a durable electronic device, for example, may assign a conversion rate of 75% of those responding they ‘definitely will buy’, yet allow for an ‘actual buyers’ rate of 15% of those responding they ‘will definitely not buy’.
Morwitz, Steckel and Gupta [2] investigated different factors that can explain variation in the strength of relationship (correlation) between purchase intentions and actual purchases (factors relating to marketing context and to methodology). Four highlights from their findings: (1) Intentions for existing products (where consumers are likely to have prior experience with the product) are more strongly correlated with purchases than for new products in the market; (2) The correlation between intentions and purchases is lower when consumers are asked at the more general level of product type or category compared with specific brand level or model / version level of the product (i.e., within category); (3) Intentions regarding durable goods (e.g., appliances) are more correlated with purchases, compared to non-durables (e.g., food & drinks) — consumers are likely to be more involved with durables and gain greater knowledge through learning about those products; (4) When consumers are asked about their intention over a shorter time horizon (e.g., 6 months) the correlation with actual purchases is higher versus a longer time horizon (e.g., 12 months or more) — consumers are likely to become aware of more information (e.g., about the product and market environment), that could make them change their minds, the longer a period that passes between the time of telling their intentions and the time for execution (note that there can be differences between types of products in this respect).
Kathryn Korostoff (CEO, Research Rockstar Training & Staffing) brings in her article (“Measuring Planned Behaviors Is Hard”, 15 May 2023) a demonstrative example of intentions-behaviours gaps (adopted from research by Ipsos). Consumers were asked first in April 2022 about intentions, how likely they were to travel and perform other leisure activities during the summer of that year; then a year later, in March 2023, respondents (in another sample — see comment below) were asked whether they had performed those activities, and how likely they were to perform them in the coming summer.
The results, as displayed in a chart in Korostoff’s article, show that the proportions of respondents intending to perform the different activities in summer 2022 are higher than those of respondents who report in 2023 performing them in the previous summer (2022); but the proportions of intenders in March 2023 again are more similar to, and at least as high as, those of intenders in April 2022. In other words, the column of ‘performers’ for each activity is ‘sandwiched’ between the two columns of ‘intenders’. For example, 50% of respondents in April 2022 said they planned to travel by plane within the US in the following summer, but just nearly 30% of respondents said in March 2023 they had taken such a flight; however, slightly more than 50% then said on the same occasion they planned to travel by plane within the US in the summer of 2023.
It is not clear that Ipsos returned to the same respondents from April 2022 again in March 2023 (as in a panel) to ask them about their own actual behaviour; that rather seems not to be the case. Hence, we cannot enquire, for instance, if the consumers forgot about their intentions from the previous year and do not calibrate their new intentions, or they are simply optimistic in the following year that they will be better able to fulfill their intentions this summer.
- Note: The wording used in questions to elicit intention — such as “intend”, “expect”, “plan”, or “likely” to do X — may not be equivalent [cf. 1]. Some expressions may be stronger inducers of intent than others (e.g., the words differ in level of commitment they imply). We may need to take more care in choosing the wording when composing the question, and later in consistently reporting the results.
Korostoff proposes three key reasons for such gaps: social desirability bias, tendency to overestimate planned actions, and the influence of context (change of plans, priorities shift). The factors intervening between intentions and behaviour can be personal, such as those suggested by Korostoff, but they can also be more market-driven (e.g., a new competitor enters the market, prices increase) or product-related (e.g., a refrigerator breaks down prematurely, a new version of a smartphone is announced), which are less in control of the consumers and can cause them to change plans. There are also social factors, like the inclination to comply with socially desirable products, services or activities, and also influence from closer people such as friends and family, and the influence of reviews and recommendations in social media networks and commercial platforms (e.g., TripAdvisor, Amazon). Korostoff offers researchers a practical suggestion, to choose specific and reasonable timeframes (e.g., likelihood of visiting a city within the next 3 months rather than within the next 12 months). Choosing the shorter timeframe may yield lower percentages for positive intentions but can narrow the gap between the planned and actual behaviours. This recommendation agrees with the finding of Morwitz et al. [2] cited above.
It is noted that discussion on intentions pertains to purchase incidence (for a given product category or brand) and does not deal with related issues in forecasting, such as the number of units to be purchased, expected amount of expenditure, or choice among alternative brands and specific product variants. It is also rightly argued that demand forecasting has to incorporate at some point (e.g., through NPD research process) variability in the price of the product.
Intentions measures are an important and useful tool in predicting purchase behaviour and forecasting demand. Yet, care has to be taken when choosing the type of intentions question by considering its implications. Researchers further need to account for factors (e.g., related to market, product, consumer) that can cause a gap or mismatch between the intentions expressed and actual purchase behaviour, and subsequently apply adjustments to reduce the gap. Predictive marketing is after all an intriguing and engaging mission for marketing researchers.
References:
[1] Methods for Forecasting from Intentions Data; Vicki G. Morwitz, 2001; in Principles of Forecasting, J.S. Amstrong (ed.)[pp. 33-55], Boston, MA: Springer.
[2] When Do Purchase Intentions Predict Sales? Vicki G. Morwitz, Joel H. Steckel, & Alok Gupta, 2007; International Journal of Forecasting, 23, pp. 347-364.