Abstract
This paper examines differences in attitudes towards electricity fee plans when information is provided on electricity bills based on past electricity consumption. We conducted randomized controlled trial stated preference (SP) and revealed preference (RP) experiments on the choice of electricity rates before and after liberalization. In the SP experiment, we measured participants' valuations of their electricity pricing plans. We found that providing information about the participants' benefit from switching diminished the tendency towards overconfidence. The valuation decreases substantially when information shows that a loss will be incurred from switching. The results of the RP and SP experiments differ. We found that the selection was not changed in the RP experiment even when providing information that a loss would be incurred.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
In the literature, some studies on switching electricity companies and contract plans have been conducted, such as Ek and Söderholm (2008), He and Reiner (2017), and Shin and Managi (2017). He and Reiner (2017) show that consumers’ attitudes towards the costs and benefits of switching affect switching behaviour. Ek and Söderholm (2008) show that the switching rate tends to be higher for consumers who obtain significant benefits from switching, whereas those with small benefits are less likely to switch. Shin and Managi (2017) show that the higher the intention to switch before liberalization, the higher the switching rate after liberalization.
For example, type \({\mathrm{s}}_{\mathrm{i}}=\mathrm{W}\) represents night consumption, and type \({\mathrm{s}}_{\mathrm{i}}=L\) represents day consumption. The peak consumption of large industrial and commercial customers occurs in the daytime, so the framework of this paper is realistic.
The following relationship holds for the payment of FLAT \(\left({\mathrm{e}}^{\mathrm{FLAT}}\right)\) and the consumer’s payment of \({\mathrm{s}}_{\mathrm{i}}=\mathrm{W}\)(\({{\mathrm{e}}_{\mathrm{h}}}^{\mathrm{TOU}}\)).
$$\begin{array}{c}{\mathrm{e}}^{\mathrm{FLAT}}-{{\mathrm{e}}_{\mathrm{h}}}^{\mathrm{TOU}}=\mathrm{p}\left(\mathrm{h}+\mathrm{l}\right)-\left({\mathrm{q}}_{\mathrm{l}}\mathrm{h}+{\mathrm{q}}_{\mathrm{h}}\mathrm{l}\right)\\ =\frac{1}{2}\left({q}_{h}+{q}_{i}\right)\left(h+1\right)-\left({\mathrm{q}}_{1}\mathrm{h}+{\mathrm{q}}_{\mathrm{h}}1\right)\\ =\frac{1}{2}\left({q}_{h}-{q}_{l}\right)\left(h-l\right)>0\end{array}$$If no information is given, then consumer \(\mathrm{i}\) considers her type as random, so we treat this case as \( \lambda = {\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 2}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{$2$}} \) .
Table 10 in Appendix 1 shows the socioeconomic attributes of the participants in each group. The average income ranged from 7.477 million yen to 7.587 million yen, which means that participants had a higher average income than Japan’s national average. For this reason, it can be understood that the participants in the current survey were comparatively wealthy, even for Japan. For this reason, it is important to be careful regarding the external validity of the results of our analysis.
Note that among the participants in the experiment, 33 had defects in HEMS data because of technical errors.
The balance check of the socioeconomic attributes of the consumer is detailed in Appendix 1. Regarding the number of people in the household, a statistically significant difference was observed for one portion at the 5% level; however, balance was achieved for other variables, and we also broadly succeeded in the randomization of socioeconomic variables.
The normal distribution is a natural choice, but when estimating the WTP, we need to restrict the cost coefficient to be negative. Additionally, if the parameter takes a value close to zero, the WTP takes an extremely large value. An approach often used to specify the sign of a parameter is to assume a lognormal distribution. However, the lognormal distribution has a long and fat tail, which causes an empirical disturbance for estimating WTP (Train, 2009).
Therefore, in estimating the selection probability, we replace the attribute value of “monthly electricity charges” with 1 if it is “the same as the present,” 0.9 if it is “10% decrease,” and 0.8 if it is “20% decrease.”.
Here, we do not compare the differences between the coefficients of the treatment group and the control group but, instead, report the coefficients obtained from each model. Because the parameters obtained using the mixed logit model featured massively different scale parameters for each model, it was not possible to perform comparisons between models. For this reason, for Tables 4 and after, the WTP, which is defined by the ratios of the coefficients, is calculated to compare results without the impact of scale parameters.
Below, regardless of whether or not those participants who have received an intervention had actually read the RECAP information, the overall effect (intention-treatment effect) of being assigned to the group that received intervention is estimated. When actually implementing a similar policy, it can be considered meaningful to estimate the ITT result because it may be assumed that some people will carefully read the information and others will not.
As robustness checks, we also use other models, such as the linear probability model and the logit model. When estimates were made using the linear probability and logit models, the results showed a similar trend to the results obtained using a probit model.
References
Agency for Natural Resources and Energy. (2017). Progress of full-scale liberalization of electricity retail, Electricity and Gas Basic Policy Subcommittee (Held on July 7, 2017). Last access September 3, 2019. https://www.meti.go.jp/shingikai/enecho/denryoku_gas/denryoku_gas/pdf/004_03_00.pdf
Agency for Natural Resources and Energy. (2018). Progress of full-scale liberalization of electricity retail, Electricity and Gas Basic Policy Subcommittee (Held on December 19, 2018). Last access September 3, 2019. https://www.meti.go.jp/shingikai/enecho/denryoku_gas/denryoku_gas/pdf/014_04_00.pdf
Bhat C (2001) Quasi-random maximum simulated likelihood estimation of the mixed multinomial logit model. Transp Res B 35:677–693
Barber BM, Odean T (2001) Boys will be boys: Gender, overconfidence, and common stock investment. Q J Econ 116(1):261–292
Bertrand M, Morse A (2011) Information disclosure, cognitive biases, and payday borrowing. J Financ 66(6):1865–1893
Borenstein S (2009) To what electricity price do consumers respond? Residential demand elasticity under increasing-block pricing. http://faculty.haas.berkeley.edu/borenste/download/NBER_SI_2009.pdf. Accessed 23 Feb 2021
Brown TC, Champ PA, Bishop RC, McCollum DW (1996) Which response format reveals the truth about donations to a public good? Land Economics, 152–166
Carson RT, Flores NE, Martin KM, Wright JL (1996) Contingent valuation and revealed preference methodologies: comparing the estimates for quasi-public goods. Land economics 80–99
Chetty R, Looney A, Kroft K (2009) Salience and taxation: Theory and evidence. American Economic Review 99(4):1145–1177
Clark J, Friesen L (2008) Overconfidence in forecasts of own performance: An experimental study. Econ J 119(534):229–251
Eil D, Rao JM (2011) The good news-bad news effect: asymmetric processing of objective information about yourself. Am Econ J Microecon 3(2):114–138
Ek K, Söderholm P (2008) Households’ switching behavior between electricity suppliers in Sweden. Utilities Policy 16(4):254–261
Ertac S (2011) Does self-relevance affect information processing? Experimental evidence on the response to performance and non-performance feedback. J Econ Behav Organ 80(3):532–545
Finkelstein A (2009) E-ztax: Tax salience and tax rates. Q J Econ 124(3):969–1010
Giné X, Mazer RK (2022) Financial (dis-) information: Evidence from a multi-country audit study. J Public Econ 208:104618
Hartman RS, Doane MJ, Woo CK (1991) Consumer rationality and the status quo. Q J Econ 106(1):141–162
He X, Reiner D (2017) Why consumers switch energy suppliers: the role of individual attitudes. The Energy Journal 38(6)
Heger SA, Papageorge NW (2018) We should totally open a restaurant: How optimism and overconfidence affect beliefs. J Econ Psychol 67:177–190
Hortaçsu A, Madanizadeh SA, Puller SL (2017) Power to choose? An analysis of consumer inertia in the residential electricity market. Am Econ J Econ Pol 9(4):192–226
Ito K (2014) Do consumers respond to marginal or average price? Evidence from nonlinear electricity pricing. American Economic Review 104(2):537–563
Kamenica E, Mullainathan S, Thaler R (2011) Helping consumers know themselves. American Economic Review 101(3):417–422
Kling JR, Mullainathan S, Shafir E, Vermeulen LC, Wrobel MV (2012) Comparison friction: Experimental evidence from Medicare drug plans. Q J Econ 127(1):199–235
Mobius MM, Niederle M, Niehaus P, Rosenblat TS (2022) Managing self-confidence: Theory and experimental evidence. Manag Sci (forthcoming)
Murakami K, Ida T, Tanaka M, Friedman L (2015) Consumers’ willingness to pay for renewable and nuclear energy: A comparative analysis between the US and Japan. Energy Economics 50:178–189
Seip K, Strand J (1992) Willingness to pay for environmental goods in Norway: A contingent valuation study with real payment. Environ Resource Econ 2(1):91–106
Sexton S (2015) Automatic bill payment and salience effects: Evidence from electricity consumption. Rev Econ Stat 97(2):229–241
Sharot T, Garrett N (2016) Forming beliefs: Why valence matters. Trends Cogn Sci 20(1):25–33
Shin KJ, Managi S (2017) Liberalization of a retail electricity market: Consumer satisfaction and household switching behavior in Japan. Energy Policy 110:675–685
Thaler RH, Sunstein CR (2009) Nudge: Improving decisions about health, wealth, and happiness. Penguin
Thunström L, Nordström J, Shogren JF (2015) Certainty and overconfidence in future preferences for food. J Econ Psychol 51:101–113
Train KE (2009) Discrete choice methods with simulation. Cambridge University Press
Author information
Authors and Affiliations
Corresponding authors
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix 1: Balance check of household attributes
Table 10 shows the average and standard deviation of participant attributes by group and the number of respondents. The third column of the table shows the difference between the average of the control group and the treatment group and the standard error. First, the proportion of females was 8.3% for the control group and 9.6% for the treatment group. The difference is 1.3% and not statistically significant (p value = 0.431). Next, regarding the working conditions of the participants, the dummy variable equals 1 when employed (including part-time) and 0 when not. The mean is 55.0% in the control group and 52.5% in the treatment group. The difference is 2.6% and not statistically significant (p value = 0.27). Regarding annual household income, the average value in the control group is 7,587.4 thousand yen ($68,980), and the average in the treatment group is 7,477.5 thousand yen ($67,980). The difference is 109.9 thousand yen ($999) and not statistically significant (p value = 0.557). Next, regarding the characteristics of the home, we first consider whether it is a detached home or an apartment. Here, a value of 1 is assigned for single-family homes, while multifamily homes take a value of 0. A total of 87.7% of the control group and 85.9% of the treatment group lived in multifamily homes. The difference is 1.8% and not statistically significant (p value = 0.351). Next, we examine whether the home is owned or rented. Here, the variable is coded as 1 for an owned home and 0 for a rental. The mean value was 99.6% for the control group and 99.8% for the treatment group. The difference is 0.2% and not statistically significant (p value = 0.54). Finally, the number of people in the household is 3.24 for the control group and 3.06 for the treatment group. The difference is 0.17 and statistically significant (p value = 0.020).
Table 11 investigates the impact of the covariates on whether the intervention is received using a linear probability model. Here, the presence or absence of the allocation of intervention (intervention = 1, no intervention = 0) is used as an explanatory variable, while factors such as energy consumption and socioeconomic data such as gender are used as explanatory variables.
The results revealed that there were no significant differences in the attributes of energy consumption, gender, employment rate, household income, whether or not individuals were living in a detached house, or whether or not they owned their homes, as shown in Table 10. Regarding the number of people in the household, a statistically significant difference was observed. It was also confirmed that the same result was found for the probit and logit models.
Appendix 2: Analysis of attrition
Here, we analyse the attrition between the SP and RP experiments. Table 12 shows the attrition rates. First, the attrition rate from the SP experiment was 7% in the control group and 8% in the treatment group. We observe no significant difference in attrition rates between the two groups (t value = 0.88). In addition, the attrition rate of the RP experiment is approximately 34% in both the control group and the treatment group, and again, no significance is observed (t value = 0.09).
In Table 13, in relation to individual attrition in the SP and RP, an analysis via a linear regression model is reported, with each attribute as an explanatory variable. A binary explained variable was used, with not participating in the SP and RP experiments being assigned 1 and a value of 0 assigned otherwise. Moreover, for both the SP and RP, the explanatory variables of the RECAP dummy, which show whether intervention has occurred, include daily energy consumption and the socioeconomic variables of the gender dummy, employment status dummy, household income, detached house dummy, home ownership dummy and number of people in the household were used. In addition, in this analysis, the intersection of the RECAP dummy and each variable attribute is included. With this approach, at the same time we estimate the impact that each attribute has on attrition, we also confirm the differences in how attributes are applied between the different groups using the intervention terms.
First, looking at the SP, there were no attribute variables that had a significant impact on attrition from experiments. In addition, no significant difference was observed regarding the impact of attribute variables on rates of attrition between groups. Next, regarding the RP data, when looking at the influence of attributes on attrition levels from the experiment, a statistically significant difference was found regarding living in a detached home and home ownership (significance level of 5%). However, there was no significant difference regarding the impact that variable attributes had on attrition between groups.
From the above, regarding the SP experiment, there was no difference in attributes among those who participated and those who dropped out and no difference based on whether there was intervention. Conversely, for the RP experiment, there was a difference observed in the attributes between those who dropped out of the experiment and those who participated in it. However, while there were differences in attributes regarding the rate of attrition, because there was no difference based on whether there was intervention, we can conclude that no bias arose from self-selection in the ultimate estimates of the intervention effects.
Appendix 3: Analysis of SP and RP including covariates
Here, the effects of intervention on both the SP and RP are estimated, inclusive of socioeconomic attributes. As with the analysis reported in Table 5 regarding the SP, the WTP of TOU gained from the SP experiment is set as an explanatory variable. Regarding the RP, as in the analysis reported in Table 7, a binary variable is used as an explanatory variable to determine whether or not people switched to the TOU payment plan after the liberalization of electricity. In addition, the explanatory variables of both SP and RP include the RECAP dummy, which shows whether intervention was received or not, as well as the socioeconomic factors of the gender dummy, the employment status dummy, household income, the detached house dummy, the house ownership dummy and the number of people in the household, as shown in Table 10. Using these variables, estimates were made using OLS for the SP. In addition, estimation was performed using the probit model for the RP.
These results are shown in Table 14. For the RP results, the marginal effects are shown in the table. First, regarding the SP, the intervention effect from RECAP is -101.19 yen, which is statistically significant at the 5% level. There is no significant difference (p value = 0.1237) in the estimated intervention effects (-112.84 yen) in Table 5, where the socioeconomic variables are not included in the explanatory variables. Regarding RP, the estimated intervention effect was -0.025, but this was not a statistically significant result. No significant difference was seen (p value = 0.9039) in the estimated intervention effects (-0.026) when the explanatory variables in Table 7 regarding socioeconomic variables were not included.
Moreover, looking at each socioeconomic variable individually, there was no significant impact on WTP or rates of switching based on gender, employment status, home ownership or detached home ownership. Conversely, regarding family income, it was discovered that regarding the SP, when income increased by 10,000 yen, WTP increased by 0.163 yen (significance level 5%). With the RP, a similarly significant increase in the rate of selection was observed (significance level 5%). In addition, it was discovered that when the number of people in the household increased by one, the WTP increased by 38.35 yen (significance level 5%). However, for RP, the coefficient for the number of household members was not significant.
From the above, in this experiment, as shown in Appendix 1, the balance was lost between the control group and the treatment group solely due to the number of people in the household. However, when socioeconomic factors were controlled for, as well as when they were not controlled for, no significant gap was observed in the estimated results of the intervention effects; therefore, it can be considered that the influence of the difference in attributes between groups is small.
Rights and permissions
About this article
Cite this article
Ishihara, T., Ida, T. The Effect of Information Provision on Stated and Revealed Preferences: A Field Experiment on the Choice of Power Tariffs Before and After Japanese Retail Electricity Liberalization. Environ Resource Econ 82, 573–599 (2022). https://doi.org/10.1007/s10640-022-00667-0
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10640-022-00667-0
Keywords
- Randomized controlled trial (RCT)
- Stated preference
- Revealed preference
- Information provision
- Power tariff
- Overconfidence