We know that people don't always do what they had intended to do. Yet, self-reported intentions continue to be used widely in academic and commercial research because they represent easy-to-collect proxies of behavior.
For example, most academic studies of satisfaction use consumers' intentions to repurchase as the dependent variable, and most companies rely on consumers' purchase intentions to forecast their adoption of new products (or the repeat purchase of existing ones).
Many studies have been conducted to improve the ability to forecast behavior from intentions. In practice, the studies adjust the intention scores by analyzing the actual purchase behavior of consumers whose purchase intentions have been measured previously. For example, the popular ACNielsen BASES model forecasts aggregate purchase rates by applying conversion rates to measured purchase intentions (e.g., it assumes that 75% of consumers who checked the top purchase-intentions box will actually purchase the product). To obtain these conversion rates, BASES uses previous studies that measured the purchase intentions of consumers and then tracked their actual purchases.
The problem: Measuring intentions makes consumers more likely to follow their intentions!
A problem of these studies is that part of the association between a consumer's intentions and behavior may be caused by the very measurement of intentions, a phenomenon called "self-generated validity." In other words: Answering a purchase intention question makes consumers more likely to remember their original intentions and to follow them at the time of the purchase.
In contrast, consumers whose intentions were not measured are more likely to be influenced by their mood of the moment, sales promotions, or any other factor unrelated to their original intentions. This raises the question of what is the true association between intentions and behavior: i.e., among consumers whose intentions have not been previously measured.
The difficulty of the task is that we need to estimate intentions for those consumers that did not participate in the survey. In other words, we are wondering what the answer is when we don't ask the question!
In our research, we provide a very simple method to do that. The idea is to measure behavior for a sample of consumers who will participate in the purchase-intention survey as well as for a control sample of identical consumers who will not participate in the survey. After that, we need to run three simple regressions:
- First, we run a regression of the self-reported intentions of surveyed consumers on the demographic and behavioral data of these consumers.
- Second, we use the results of the first regression to predict the intentions of both surveyed and non-surveyed consumers. To be able to do that, we have to assume that both samples are identical (i.e., people were randomly assigned to the survey or control groups).
- Third, we estimate the correlation between these new "predicted" intentions and behavior in each group.
Intentions do not predict behavior as well as we thought
We applied this procedure to three large-scale field studies. The first measured the repeat-purchase intentions, purchases, and firm profitability of existing customers of a French online grocer. The second and third studies measured the intentions and purchases of automobiles and personal computers of two large samples representative of the entire US population.
On average, the correlation between predicted intentions and purchasing was 58% greater among surveyed consumers than among similar non-surveyed consumers. We also found that a one-point difference in predicted purchase intentions (measured on a five-point scale) in the grocery study lead to a €52.71 gain in customer profitability when intentions are measured—but was worth only €23.95 when intentions are not measured.
In addition, we found that measuring intentions increased purchase behavior among consumers with positive intentions but decreased purchase behavior among consumers with negative intentions. This shows that the measurement effects occurred because the surveys made consumers more likely to remember—and then use—their pre-existing intentions.
The surveys did not simply make consumers like the product or the company better, or encouraged them to say "yes" out of politeness, regardless of their true intentions. Had this been the case, we would have found higher purchasing even among consumers with negative intentions.
Implications: Beware of intentions
Our results show that the predictive power of purchase intentions is significantly weaker and less reliable than is commonly measured. One important implication is that commonplace procedures and models in which the intentions-behavior relationship is measured using the same sample of consumers overestimate the association between intentions and behavior.
As a result, these studies lead marketers to focus too much on consumers with high purchase intentions. In reality, marketers should consider enlarging their customer target to include also consumers with neutral or even negative intentions, as they are more likely to purchase than what they claim. Another implication is that when the concepts tested and the purchase intentions are positive, existing models overstate aggregate purchase probabilities.
More generally, we argue that any studies that go beyond pure descriptions and intend to examine the association between concepts should test for self-generated validity effects. This applies not only to intentions, but also to measures of beliefs, attitudes, or satisfaction.
Fortunately, that is relatively easy to do: Simply keep a control sample of consumers for whom you will not measure the concept and use the method described in this research to see if they behave the same way than consumers who answered the questions.
Note: The original of this article appeared in the Journal of Marketing, 69 (2), 1-14, "Do Intentions Really Predict Behavior? Self-Generated Validity Effects in Survey Research." It is reprinted here with permission of the Association for Consumer Research.
