Hypothetical bias and framing effect in the valuation of private consumer goods

Understanding the mechanisms that influence consumers’ decisions and their perception of the value of products and services is a subject of intense consideration in numerous fields of science, including economics, psychology, marketing, and management. Researchers and business practitioners are continuously trying to evaluate the determinants of the valuation of various goods by consumers and to find methods and models that help to appropriately establish prices.

In this paper, I examined the influence of selected psychological factors, called behavioural effects, on the valuation of private consumer goods. As such, I focussed on two well-known phenomena: hypothetical bias – the overestimation of somebody’s willingness to pay (WTP) in declarative situations – and the framing effect – differences in decisions made depending on how the information is presented. I investigated these effects in a laboratory experiment with private goods and thus verified their impact on participants’ WTP. The interaction between the aforementioned phenomena was also examined, which is a novel concept in the literature on the subject; considering how the interaction between the framing effect and hypothetical bias affects WTP values appears to be critical for both practical and theoretical reasons. The research methods using only hypothetical or declarative methods for eliciting consumers’ WTP are very popular. The literature noted that hypothetical WTP values exceed actual ones; however, as the size of the hypothetical bias is not constant, we cannot solve this problem by calibrating participants’ responses. In this case, we can suppose that the quality of the hypothetical data is inferior to data from real purchasing situations. In this paper, I seek to verify this conjecture, scrutinising whether the tendency to follow certain premises contrary to the paradigm of rationality (such as the framing effect) is stronger when the reported WTP has no financial consequences for the participants. On the other hand, this paper may also contribute to the framing effect literature; an interaction between the studied effects would suggest that the framing effect may simply be a result of insufficient incentives to reveal the actual WTP.

The results of the study may have both theoretical and practical implications. First, this paper will supplement and broaden the existing knowledge of the determinants of the valuation of consumer goods, as well as improve current research methods for eliciting the market prices. As using the framing effect (especially, the positive attribute framing) is remarkably common in advertising messages and slogans, the research will also facilitate recommendations within marketing sciences.

The first behavioural effect relevant to my study is the hypothetical bias. Individuals overstate their valuation in declarative research in comparison to their actual WTP, as confirmed by meta-analyses such as Murphy and Stevens (2004). In the most recent meta-analyses, covering 77 studies, Foster and Burrows (2017) ascertained that the median hypothetical WTPs exceed values observed in situations with real transactions (or incentivised trials) by as much as 39%.

In the literature, many different research methods were used to obtain participants’ WTP values. The methods used to elicit hypothetical values include choice experiments (Moser, Raffaelli, & Notaro, 2014), direct elicitation (Doyon, Saulais, Ruffieux, & Bweli, 2015), and declarative Vickrey auctions (List, 2001). To obtain the actual values, the authors largely used auctions (List, 2003) and the BDM (Becker, DeGroot, & Marschak, 1964) procedure (Boyce, Brown, McClelland, Peterson, & Schulze, 1989); however, some studies related to hypothetical bias only observed purchasing decisions instead of eliciting subjects’ specific WTPs (Blumenschein, Johannesson, Blomquist, Liljas, & O’Conor, 1998). The general conclusion from the extant literature is that the type of valuation method can moderate the hypothetical bias; choice-based elicitation methods, in particular, may reduce it (Murphy, Allen, Stevens, & Weatherhead, 2005).

Several studies have also indicated that a number of specific tools are able to reduce hypothetical bias. These include both cheap talk (Doyon, Saulais, Ruffieux, & Bweli, 2015; List, 2001) and real talk (Alfnes, Yue, & Jensen, 2010), as well as various calibration techniques. Some researchers have also found that the usage of student samples may contribute to the bias (Murphy, Allen, Stevens, & Weatherhead, 2005) and that hypothetical bias is weaker in experiments with private goods compared to those with public goods (List & Gallet, 2001).

The second of the selected behavioural effects analysed in the study is the framing effect (see Piñon & Gambara, 2005 for a review). This phenomenon refers to the dependence of the decision made on the formulation of the decision problem. This concept was introduced by Tversky and Kahneman (1981), who linked it to their prospect theory (perceiving the effects of decisions in terms of profits vs. losses). Levin et al. (1998) distinguished different types of framing effects: the risky choice framing effect, the attribute framing effect (associated with the presentation of the attributes of the good), and the goal framing effect (differentiating the decisions of the respondents depending on the way in which the effects of the action are presented).

From the viewpoint of the paper’s subject, the results of previous research on the attribute framing effect are of paramount importance. Among them is the well-known study of Levin and Gaeth (1988), in which the designation of beef with a 75% lean label influenced a higher taste rating over the same beef described as 25% fat; similar results were provided by other studies related to the presentation of product features. Levin et al. (1998) proposed an explanation of this phenomenon: “The positive labeling of an attribute leads to an encoding of the information that tends to evoke favorable associations in memory, whereas the negative labeling of the same attribute is likely to cause an encoding that evokes unfavorable associations” (p. 164).

Several studies have also indicated that a number of respondent characteristics may affect the power of the framing effect: for example, Braun, Gaeth, and Levin (1997) demonstrated that women are more sensitive to attribute framing than men when the chosen attribute is the percentage of fat in chocolate (20% fat or 80% fat free). Hardisty, Johnson, and Weber (2010) showed that attribute framing of a carbon fee as either a tax (negative framing) or as a cost offset (positive framing) affected Independents and Republicans but had a lesser impact on Democrats. In light of these results, another important factor to examine is the influence of respondents’ sociodemographic characteristics on the operation of the framing effect in the valuation context.

Only a small fraction of the framing literature is directly related to eliciting WTP. Most studies concern either judgements or assessments. Moreover, it should be highlighted that they typically involve solely hypothetical choices.

There is some research on the framing effect comparing real outcomes with hypothetical ones; however, what should be stressed is that none of them are related to private consumer goods and their valuation. Typically, they concern gambling (Kühberger, Schulte-Mecklenbeck, & Perner, 2002; Levin, Chapman, & Johnson, 1988; Wiseman & Levin, 1996) or time allocation decisions (Paese, 1995; Wiseman & Levin, 1996); moreover, no studies have found any significant differences in participants’ choices.

I formulated the following research hypotheses:

H1: Hypothetical bias influences the valuation of the product

This hypothesis is based on a broad literature. I expect that participants who make hypothetical, declarative decisions indicate higher WTP for a given product than those making decisions with actual financial consequences.

H2: The framing effect influences the valuation of the product

In order to verify this hypothesis, I will use one of the types of framing, which consists of differentiating the ways of presenting the product’s attributes. I expect that participants who have been exposed to positive attribute framing have a higher WTP than subjects who have been exposed to negative framing.

H3: Interactions between hypothetical bias and the framing effect

I assume that the framing effect will be stronger in the case of hypothetical decisions than in actual purchasing decisions. In other words, I expect that negative attribute framing will reduce hypothetical bias compared to positive framing.

To verify the hypotheses, I conducted a laboratory experiment. The participants were asked to reveal their WTP for a tube of mascara. I chose this good because it is a well-known everyday product that is diversified by price, model, and brand. The price for mascara in Poland ranges from 10 PLN to more than 200 PLN. I selected the Cover Girl Lash Blast Clump Crusher mascara because it is inaccessible in cosmetic shops in Poland (and is rarely found online); so I expected that participants would not know the market price (57.40 PLN or approximately 13.37 EUR in an online shop in November 2016). Post-experiment questionnaires confirmed my assumptions: only a very small number of participants were acquainted with the brand or with the model.

My experiment was paper and pencil. Four different treatments were used in this experiment in a 2 x 2 design: RealN (real transactions, negative attribute framing), RealP (real transactions, positive attribute framing), HypoN (hypothetical valuation, negative attribute framing), and HypoP (hypothetical valuation, positive attribute framing). Participants were randomly assigned to either a real or hypothetical condition at the session level, whereas negative or positive attribute framing was randomly assigned to each subject within a session.

In treatments with real transactions, at the beginning, I informed participants of the rules of the experiment (see Appendix). Next, they took part in a BDM procedure; this method is regarded to be correct in terms of incentive compatibility (Kagel, 1995). In this procedure, I asked participants to give the maximum price (s) they would be willing to pay for the presented product. Thereafter, we drew the transaction price (p) from a pre-specified distribution. Participants were not informed of the range of the distribution to avoid anchoring (Bohm, Lindén, & Sonnegård, 1997). If the participant’s offer was higher than or equal to the selected price, the participant was required to buy the product at price p. If the offer was lower than the selected price, the transaction was not executed. The weakly dominant strategy in this procedure is to state the true WTP (Wertenbroch & Skiera, 2002).

Participants were carefully informed of the principles of the procedure used, both orally and in writing. I subsequently showed them the mascara and presented the additional information about the product and framing sentence. Afterwards, I distributed the valuation questionnaires with the following request: Give the maximum price that you are willing to pay for CoverGirl LashBlast Clump Crusher Mascara (Please enter a specific amount in PLN). I then randomly chose the transaction price and the participants who had the opportunity to buy the product. In the end, a number of actual transactions were executed.

In treatments with a hypothetical valuation (HypoLow and HypoHi), the scheme was similar, but I used the direct declarative method of eliciting participants’ WTP for the product instead of the BDM procedure. I informed the subjects that their valuation was purely declarative. In valuation questionnaires, participants were asked: Give the maximum price that you would be willing to pay for CoverGirl LashBlast Clump Crusher Mascara. (Please enter a specific amount in PLN)

In order to test the framing effect, I compared the positive and negative framing. In the HypoP and RealP treatments, the positive framing was formulated as follows: “As many as 71% of users would buy this product again (information from the makeupalley. com)”. Instead, the negative framing in the HypoN and RealN treatments was as follows: “Only 29% of users would not buy this product again (information from the makeupalley.com)”

In the end, in all treatments, the participants were asked to complete the post-experimental questionnaire concerning their shopping habits and consumer preferences regarding cosmetics, as well as their sociodemographic characteristics.

It took about 15 minutes for each session to be conducted. Participants were given both oral and written instructions. The experiment was conducted at the Faculty of Economic Sciences of the University of Warsaw. In total, 167 female students took part in my experiment; their mean age was 20 years. A typical participant was an unemployed student in a good financial situation.

The mean WTP for the product in the overall sample was equal to 25.47 PLN, the median was 25 PLN, and the standard deviation was 16.46 PLN. I began my analysis by comparing the WTP values in each treatment. Table 1 and Figure 1 show the values of WTP by treatment.

Fig. 1

When analysing Figure 1, we are able to notice that the mean WTP is considerably lower in the real treatments than in the hypothetical ones, while the anchor made little difference. To verify these observations, I used non-parametric Mann–Whitney U tests; first, we compared treatments with hypothetical valuations and with real transactions (separately for the positive and negative attribute framing) to verify the hypothetical bias. Differences between the WTP values in HypoN and RealN treatments were statistically significant (p=0.0000), as well as in HypoP and RealP treatments (p=0.0000); therefore, we can confirm the hypothetical bias. Next, I compared treatments with positive and negative framing (separately for the groups with hypothetical valuations and real transactions). In the first pair of treatments – HypoN and HypoP – differences between WTPs were not statistically significant (p=0.2430). Similarly, in RealN and RealP treatments, we did not observe statistically significant differences (p=0.5567). In light of these results, we can conclude that the framing effect had no impact on the valuation of the presented mascara.

In the next step, I estimated a simple ordinary least-squares (OLS) regression model to identify factors that impacted the valuation of the mascara. The dependent variable in this model is the participant’s WTP value. Most of the independent variables are extracted from the post-experimental questionnaire, and two of them represent experimental conditions (hypothetical and framing_pos). In Table 2, four specifications of the model are presented; Model M4 is the final form of the OLS.

In the final form of the OLS model, we obtained six statistically significant variables:

It is worth mentioning that the interaction between the two most important variables, framing_pos (1 – for positive framing, 0 – for negative framing) and hypothetical (1 for declarative (hypothetical) valuation, 0 – for real transactions) was not statistically significant.

The RESET test showed that the functional form of the model was correct (F(3, 150)=2.22; p=0.0878). The results are consistent with those of non-parametric tests reported previously; indeed, the hypothetical bias influences the valuation of the product – in a hypothetical valuation, WTP is higher than a BDM valuation; however, the framing has no statistically significant effect.

To summarise, the results of the current study demonstrated that hypothetical bias influences the valuation of goods. Participants who valued mascara hypothetically had a higher WTP than the others by 12.34 PLN; however, we cannot observe the impact of the framing effect on WTP for the mascara, nor the interaction between hypothetical bias and the framing effect.

I confirmed Hypothesis 1, that hypothetical bias influences the valuation of the product. The study demonstrated that the hypothetical bias determines the perception of goods and influences their valuation. Participants overstated their actual WTPs in hypothetical situations. However, when using declarative methods, we are not able to discern participants’ actual preferences, only their hypothetical ones; thus functioning primarily in the symbolic sphere. The experimenter demand effect or social desirability bias may also be the source of the inflated WTP values.

I did not confirm Hypothesis 2, that the framing effect influences the valuation of the product. Similarly, I rejected Hypothesis 3, that hypothetical bias and the framing effect interact. Of course, the reason for the latter may be that the framing per se was too weak and so did not affect the valuation; this outcome may be a random incident because of the relatively small sample. The usage of a mild frame or a mismatch between the type/formulation of the framing information and the sample may also play a role.

Nevertheless, in light of these findings, we cannot conclude that hypothetical data are of lower quality than actual data. Hypothetical results are shifted towards higher values but remain internally consistent. In this sense, there is no reason to question the validity of declarative research methods. It is worth remembering, however, that I investigated only one manifestation of supposed low quality of hypothetical data, so further research should be pursued.