Decades of research on affective forecasting have shown a persistent intensity bias—a strong tendency by which people overestimate their future hedonic response for positive events and underestimate it for negatives one. While previous research has provided answers on the isolated impact of various individual or contextual factors, this study is original in that it brings them together to determine which ones most influence the inaccuracy of affective forecasting. Participants were asked to predict their emotional satisfaction for a personal life event, the course (positive or negative) and date of which were already known. First, the results support previous research by showing that affective predictions are highly associated with people’s affective experience. Moreover, multiple regression showed that among the individual and contextual factors previously reported to be in relation with affective forecasting inaccuracy, only the valence of the event could explain inaccuracy of forecasting. According to a growing body of literature, these findings point out a tendency to underestimate the intensity of the affect predicted both for negative and positive, with a stronger underestimation for negative events: the negative valence effect.
Forecasting how we are going to feel in the future; that is what
Despite the adaptive role of our exceptional ability to mentally travel in time, several lines of evidence have showed that affective forecasting is subjected to many errors (e.g.,
Although the intensity bias is considered as a robust phenomenon, recent studies have pointed out some methodological problems behind the studies shedding light upon it, and others have demonstrated that both dispositional and contextual factors could modulate it (
Recently, some authors have raised concerns about the usual means of measuring and interpreting affective forecasting inaccuracies, challenging the significance and validity of the intensity bias (
A main concern that directly challenges the intensity bias is relative to a procedural artifact. Thus, the inaccuracy or the accuracy of our affective forecasting could be considered as being two facets of the same reality according to the choice of the data analyses (
In order to better understand the conditions of occurrence of the intensity bias, a group of studies explored the impact of several factors on people’s affective forecasting (e.g., personality traits, event valence, importance or emotional intelligence). The purpose of these studies was to point out whether individual traits or contextual factors may lead individuals to overestimate or underestimate their future emotions, in other words,
First, several studies focused on individual traits reported evidence supporting the link between personality traits and affective forecasting (e.g.,
Previously, research indicated that people believe unrealistic optimism to be preferable to realism (
According to
Many studies have examined the role of the valence of the event on affective prediction through its direct or indirect “positive” or “negative” categorization. For example,
Through its direct or indirect consideration, the valence was one of the most common factors found in the literature about affective forecasting. To our knowledge, it was the only contextual factor to have been studied along with many other dispositional factors in the affective forecasting accuracy literature. Although research has shown that both positive and negative valences can impact the prediction of affect and accuracy, most showed a stronger effect for negative events (e.g.,
While dispositional traits and the valence were the first factors to be investigated to try to identify what might increase or decrease inaccuracy in affective forecasting, other contextual factors were subsequently explored. Few studies have investigated the implication of subjective importance given by participants to the event for which affective forecasting was asked (
Could the fact that we have already experienced an event improve the accuracy of forecasting? Previous studies have highlighted the relationships between memory and predictions (
While many individual and contextual explanatory factors have been investigated mainly separately in affective forecasting research, few studies—if any—have yet studied them together. With regards to theoretical contribution, to the best of our knowledge, the present study provided the first piece of evidence examining which of these factors are the strongest predictors of affective forecasting inaccuracy. Considering the two distinct ways of approaching the accuracy of affective forecasting, we propose to explore both the absolute and relative forecasting accuracy through different analyses.
The first step will be to assess the relative accuracy between the predicted and experienced scores with a correlation analysis. Following
Then, to answer our main question, we will include in the same analysis: extraversion, neuroticism, optimism and emotional intelligence as dispositional factors, and the subjective importance of the event, the valence and frequency as contextual factors to explore which of them predict inaccuracy in affective forecasting for a likely life event. According to the predictors detected by the multiple regression model, we will know which contextual or dispositional factors most impact the absolute accuracy of affective forecasting for a personal life event.
Outside of valence, there is little data on the combined impact of contextual and individual factors on affective forecasting. Therefore, no assumptions can be made about the prevalence of some predictors over others. However, given the robustness of the valence found in multi-factor research, we expect that its effect will persist alongside other factors in our analyses.
A total of 335 participants was recruited in the main hall of the university hospital. They were asked to participate in a study on positive psychology. The main investigator of the study informed them that they would be asked to answer a series of online questionnaires on a tablet for 30 minutes. After providing informed consent, demographic measures of age, gender and highest level of diploma (ranging from 1 to 7 for primary school degree, lower secondary, high school, bachelor, master, PhD and honorary degree) were asked. Additionally, the current affect was measured by the Positive and Negative Affect Schedule (PANAS;
In the present study, participants were given the freedom to predict their feelings about a personal and likely event (for additional design information, see SM1 in
In previous research we have seen that the valence of the event can be arbitrarily attributed by researchers (e.g., Christmas Day as a positive event and no Valentine's Day date as a negative event). However, two people can attribute a different valence to the very same event depending on their ethnicity, age, needs, past experiences, life goals and personal characteristics. In the present study, participants were randomly instructed to give either a positive or a negative personal future life event. Through this procedure, we ensured that the valence data were in accordance with the subjective valence of the participant.
Considering the role of the likelihood of the outcome of an event on the accuracy of affective forecasting (
Finally, the event should fulfill two timing conditions: it should be scheduled on a specific date already known at the time of answering the questionnaire and, occurring 4 weeks after predicting the affect. A tolerance of 2 days around the 4-week delay was granted in order to give the participants the opportunity to find an event that was already scheduled. Like the occurrence, event timing was therefore a fixed factor common to the entire sample. Only the valence, as a dichotomous variable, presented categories and was therefore included in our analyses as a contextual factor.
Once the personal event had been selected, the subsequent contextual measure of
Time 1- Participants were asked to perform a forecast of the level of their emotional satisfaction about the event (for detail about the choice of the affective target, see SM2 in
Time 2- A month later, the event for which affect had been predicted occurred. The experienced emotion was evaluated by phone within an 8-hour delay following the event. After having been asked if the event had occurred, participants were requested to rate their current affective state regarding the event on the same Likert’s scale. Following previous warnings found in the literature on affective forecasting about the method (see,
Past research on affective forecasting has been based on a conceptualization of accuracy as the mathematical difference between the forecasted and the actual affect (
Ninety-one percent of the 335 individuals accepted to participate. We excluded one participant who did not answer to the recall concerning experienced emotion, six for whom the event had not occurred, and four for whom the event that the participants actually experienced was not the one initially chosen. Analyses were conducted on a final sample composed of 295 individuals.
Prior to the main analysis, a preliminary analysis was conducted to determine whether the data was suitable for the analysis to be performed. Kurtosis statistics were used to examine normality assumption. According to
Variable | Range | Kurtosis | ||
---|---|---|---|---|
1. Gender | 0 - 1 | 0.45 | 0.498 | −1.97 |
2. Age | 16 - 86 | 40.83 | 15.31 | −0.901 |
3. Education level | 1 - 7 | 3.67 | 1.6 | −0.805 |
4. Positive affect | 12 - 40 | 27.76 | 4.84 | −0.188 |
5. Negative affect | 11 - 41 | 27.42 | 4.91 | 0.163 |
6. Extraversion | 1.37 - 5 | 3.22 | 0.731 | −0.445 |
7. Neuroticism | 1 - 4.75 | 2.83 | 0.823 | −0.698 |
8. Emotional intelligence | 92 - 197 | 144 | 19.08 | −0.132 |
9. Optimism | 1 - 24 | 14.4 | 4.22 | −0.059 |
10. Dispositional happiness | 8 - 28 | 19.9 | 3.72 | 0.468 |
11. Valence | −1 - 1 | 0.105 | 0.996 | −1.96 |
12. Importance | 1 - 10 | 5.98 | 2.82 | −0.997 |
13. Frequency | 0 - 3.6 | 0.937 | 0.823 | −0.143 |
14. Predicted score | 1 - 10 | 6.99 | 1.95 | 0.208 |
15. Experienced score | 1 - 10 | 7.49 | 1.97 | 0.214 |
16. Inaccuracy score | −5 - 4 | −0.51 | 0.498 | 0.698 |
Pearson correlation coefficients were estimated among all the study variables assessed in the present study.
Variable | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1. Age | − | |||||||||||||
2. Education level | −.134* | − | ||||||||||||
3. Positive affect | −.044 | −.009 | − | |||||||||||
4. Negative affect | −.034 | −.003 | .804*** | − | ||||||||||
5. Extraversion | −.004 | .019 | .146* | .054 | − | |||||||||
6. Neuroticism | −.152** | .021 | −.175** | .216*** | −.201*** | − | ||||||||
7. Emotional intelligence | .025 | .159** | .007 | −.080 | .514*** | −.534*** | − | |||||||
8. Optimism | .098 | .139* | .002 | −.066 | .299*** | −.389*** | .511*** | − | ||||||
9. Dispositional happiness | .056 | −.006 | .033 | −.087 | .318*** | −.489*** | .520*** | .535*** | − | |||||
10. Importance | .192** | −.178** | .238*** | .213*** | .107 | −.077 | −.009 | .069 | .050 | − | ||||
11. Frequency | .170** | −.004 | −.080 | −.102 | −.054 | −.002 | −.011 | −.033 | −.068 | −.103 | − | |||
12. Predicted score | .213*** | −.122* | .035 | .030 | .049 | −.152** | .103 | .140* | .190*** | .268*** | −.027 | − | ||
13. Experienced score | .164** | −.158** | .034 | .069 | .101 | −.146 | .138* | .147* | .206*** | .242*** | −.090 | .700*** | − | |
14. Inaccuracy score | .062 | .049 | .001 | −.051 | −.068 | −.006 | −.047 | −.012 | −.023 | .031 | .082 | .375*** | −.399*** | − |
*
In order to find which dispositional and contextual factors primarily impact the inaccuracy of affective forecasting, a multiple regression model was generated. This procedure allows for multicollinearity between some of the predictors to be handled (see
Dependent variable | Unstandardized β | Standardized β | Collinearity statistics |
|||
---|---|---|---|---|---|---|
Tolerance | VIF | |||||
Individual factor | ||||||
Extraversion | −0.133 | 0.143 | −0.064 | .353 | .694 | 1.441 |
Neuroticism | −0.048 | 0.134 | −0.026 | .723 | .626 | 1.597 |
Emotional intelligence | −0.003 | 0.007 | −0.032 | .703 | .469 | 2.133 |
Optimism | 0.011 | 0.026 | 0.031 | .670 | .878 | 1.139 |
SHS | −0.005 | 0.031 | −0.012 | .870 | .655 | 1.526 |
Contextual factor | ||||||
Valence | 0.527 | 0.185 | 0.173** | .005 | .907 | 1.103 |
Importance | −0.003 | 0.033 | −0.005 | .938 | .867 | 1.154 |
Frequency | 0.153 | 0.108 | 0.083 | .160 | .969 | 1.032 |
**
We performed this multiple regression analysis with the inaccuracy score as dependent variable and with extraversion, neuroticism, emotional intelligence, optimism, dispositional happiness, valence (dummy coded −1 = negative; 1 = positive), importance and frequency as predictors.
Although the multiple regression reveals the direction of the inaccuracy score according to the predictors, the analysis does not allow us to know the details of the forecasting error. By using the inaccuracy score, the slope direction alone does not allow us to determine if the participants have under- or overestimated their affect. Indeed, one needs to take into account the predicted values of the dependent variable, as a similar slope coefficient may either translate the fact that overestimation is observed for a positive event, while underestimation holds true for negative events; or the participants could have simply underestimated to a lesser degree their affect for a positive rather than for a negative event. To better understand the direction of the impact of this last significant predictor, we examined how much intensity bias differed by experimental condition of valence.
In order to evaluate the inaccuracy of participants in the absolute sense according the valence, emotional satisfaction ratings were submitted to a 2 (time: predicted vs. experienced) × 2 (valence: negative vs. positive) repeated measure ANOVA. The analysis revealed a dominant effect of time,
As expected, the present study showed that participants were both accurate and inaccurate to predict their feeling. According to
The certainty of the outcome of the events could also have influenced the accuracy of the forecasts. Historically, affective forecasting was mainly studied for events with unlikely outcomes like a housing attribution (
In the present study, the valence was the only predictor of the intensity bias when predicting the level of emotional satisfaction for a personal and certain event, meaning that subjects experienced a higher level of emotional satisfaction than they had predicted 1 month earlier, both for negative and positive events (see
Even if for negative events, the intensity bias commonly predicts an underestimation of the forecasted hedonic response, the slight underestimation of the forecasted hedonic response reported in the present study for positive events is not congruent with previous findings usually showing the opposite pattern (i.e., overestimation). However, a growing part of the forecasting literature challenges the classical intensity bias direction, and more particularly through this inversion of the direction of the error for events with a positive valence (
In the present study, underestimations of the level of predicted emotional satisfaction for negative and positive events bring additional findings in favor of a new line of studies challenging the direction of the traditionally observed forecasting error. Together, these findings indicate that the bias is less uniform than suggested previously.
Even if the underestimation of the level of emotional satisfaction is significant for both valence conditions (negative:
Before drawing any conclusions, it is necessary to put forward several limitations of the study in order to more critically interpret its results. The main strength of this study is at the same time its greatest weakness: the diversity of events used for forecasting by the participants. Although this methodological choice allowed to provide a higher variability within responses among the many studied factors, and guaranteed that subjective valence indeed went in the intended direction, it nonetheless limited the scope of our conclusions on the resulting relationships between variables. As a result, interpretation of results must be carried out with utmost caution. In the future, the use of personal life events limited to fixed predefined categories (e.g., health, family or professional events) could be a methodological compromise to more solidly back up the present data. Secondly, the procedure was thought up to limit the impact of occurrence probability of the event through the instruction to forecast only for an event for which the outcome was absolutely certain. An interesting question for future research would be to use a similar multifactorial procedure by including groups of events with different levels of probability of occurrence. Finally, following the robustness of the negative effect found in this study and throughout the literature, we suggest that future studies should include the valence as a continuous variable rather than a categorical one.
These findings suggest that among the main dispositional and contextual variables previously studied in the literature, the subjective valence of the event is the stronger predictor of inaccuracies in affective forecasting. In addition, our findings showed a stronger underestimation of the forecast for negative life events compared to positive ones which can be considered as low, representing a negative valence effect. We found an underestimation of the emotional satisfaction both for the positive and negative events, that is, a classical direction of misestimation for negative events but a reversal of direction for the misestimation of positive ones. Congruently to a growing body of literature in affective forecasting, our results challenge the direction of the traditional intensity bias. Finally, the results of the present study support that people can be both accurate in the direction of the forecast of their affect and inaccurate in the absolute intensity of their prediction according the valence of the predicted event.
For this article the following Supplementary Materials are available via the PsychArchives repository (for access see
SM1: Additional design information.
SM2: Detail about the choice of the affective target.
SM3: Metric issue in affective forecasting literature.
SM4: Discussion about the averages of emotional satisfaction for negative events.
SM5: Discussion about the stronger value of the effect observed for negative events through memory bias and motivational theories.
This research was funded by the University of Liège under Grant Number NFB-14/19, and meets its expectations in regards to research data, which are available in the Open Science Framework (OSF) public data repository.
The authors have declared that no competing interests exist.
The authors have no additional (i.e., non-financial) support to report.
Procedure performed in this study on human participants was approved by the ethical committee of the University of Liège Psychology School and in accordance with the 1964 Helsinki declaration and its later amendments. All participants included in this study gave their written informed consent.