This experiment began with thirty participants (all students, mean age 21.17 ± 1.93 years, range 18–26, 15 males and 15 females). All participants were right-handed, and native speakers of Japanese. Four participants for whom, based on their mean WTP scores, the IKEA effect did not occur (WTP for the three DIY products was not higher than or the same as the mean WTP for the three Non-DIY products), and two participants who reported typing errors were excluded from subsequent analyses. As a result, the data for 24 participants were analyzed. The remaining 24 participants confirmed in the post-experiment questionnaire that they had never purchased or assembled any of the six products used in this experiment. In addition, 23 participants had never seen any of the six products. One participant indicated that he had seen the six products, but not assembled them. However, because there was a discrepancy between the actual price and the market price predicted in the post-experiment questionnaire, it was determined that this did not affect his WTP response. Before the experiment, informed consent was obtained from all participants. Participation was voluntary, and an honorarium as well as the six IKEA products used for the experiments were given to the participants after the experiment. This study was conducted after obtaining permission based on the Declaration of Helsinki from the Ethics Committee at Chuo University.

Participants were informed in advance that they should not eat or ingest caffeine for 2 h prior to the experiment. Just before the fNIRS measurements, participants responded to two questionnaires. We conducted Edinburgh's dominant hand test and confirmed that participants were right-handed (Oldfield, 1971). The Mini International Neuropsychiatric Interview (M.I.N.I.) was used to screen out participants with psychiatric disorders (Sheehan et al., 1998) because Blumberg et al. (2003) and Aupperle et al. (2012) reported that brain activation in psychiatric patients with PTSD and bipolar disorder is more filtered than that in healthy individuals. After the study was completed, we told the participants that they would be given these six products.

In order to determine which six products would be appropriate for the current study, we conducted a preliminary questionnaire survey in June 2020 that targeted 13 out of over 1,500 IKEA products available at IKEA stores in Japan and priced at 999 yen. The preliminary questionnaire included WTP, free-entry survey of projected market prices, and product evaluation using the five-item method. Three DIY products with similar difficulty levels were selected based on the questionnaire results and their respective assembly instructions, and three Non-DIY products were selected based on the questionnaire results. The results of the preliminary questionnaire confirmed that there was no significant difference in WTP scores between the six IKEA products [F_(5,135) = 0.727, p = 0.61].

Experimental procedures are depicted in Figure 1. Before the fNIRS measurement, participants were asked to assemble the three DIY products and to inspect the three Non-DIY products for at least 1 min. We randomized and alternated the order of the DIY condition and the Non-DIY condition to eliminate order effect. Participants were then given a 10-min break followed by the WTP evaluation of the six products, which was taken along with the fNIRS measurements. The entire experiment lasted about an hour. The WTP evaluation task was conducted for each of the six products (three products in the DIY condition and three products in the Non-DIY condition) for each participant in an event-related design. The order of the products in the WTP evaluation task was randomized among the participants. For the WTP evaluation task, an experimental program was created using E-prime 2.0 (Psychology Software Tools, Pittsburgh, PA, USA). All the directions displayed during the task were written in Japanese. During the experiment, the participants were asked to follow the directions displayed on the monitor. For instructional purposes, the screen displayed “Later we will ask you how much you would be willing to pay for this item.” for 3 s and then the images of the six products were randomly displayed for 15 s each. Then, “How much would you be willing to pay for the product?” was displayed for 6 s and the participants entered their WTP amount. In addition, a gazing point was shown on the screen for 16–20 s at random. This process was repeated six times to evaluate each participant's WTP for the three Non-DIY products and the three DIY products. The WTP evaluation task took about 42 s per trial, so it took about 4.2 min total to measure the six products. Before the measurements were taken, the WTP evaluation task was practiced on three products unrelated to the experiment.

We used the Hitachi ETG-4000 to measure cerebral blood flow response using two near-infrared light sources (695 and 830 nm). We analyzed the fNIRS data using a modified Beer-Lambert law (Cope et al., 1988). This method is used to calculate the change of oxyhemoglobin (oxy-Hb) and deoxygenated hemoglobin (deoxy-Hb) signals in units of millimolar × millimeter (mM × mm) based on the intensity change of near-infrared light (Maki et al., 1995). The sampling rate was set at 10 Hz. In this experiment, we used 3 × 11 multichannel holders (Figure 2). The probe used in the experiment consisted of 17 illuminator and 16 detector alternately spaced 3 cm apart, for a total of 52-channel sites measured. The probe was mounted according to the international 10–20 system as a reference point. First, the multichannel probe holder was placed such that the detector in the middle of the lowest row corresponded to Fpz. Then, the illuminators and detectors in the lowest row were matched to the horizontal reference curve, which was determined by a straight line connecting T3-Fpz-T4 (Klem, 1999; Jurcak et al., 2007).

After measuring cerebral blood flow, the location of all the optodes and the landmarks such as Nasion, Inion, CZ, and bilateral Preauricular Reference Points were acquired using the Polhemus Patriot digitizer (Polhemus, Colchester, VT, USA). We employed the probabilistic registration to register fNIRS data to Montreal Neurological Institute (MNI) standard brain space (Tsuzuki et al., 2007; Tsuzuki and Dan, 2014). The projected spatial data were mapped to anatomical labels (Okamoto et al., 2004; Okamoto and Dan, 2005). The spatial registration data were registered with macro-anatomical labeling (Okamoto et al., 2004; Okamoto and Dan, 2005) in reference primarily to Brodmann's atlas (BA) (Rorden and Brett, 2000) and secondarily to the macro-anatomical labeling in LPBA40 (Shattuck et al., 2008).

In the first-level analysis, time series data for oxy-Hb and deoxy-Hb were analyzed using in-house MATLAB analysis tools developed in the Applied Cognitive Neuroscience Laboratory at Chuo University (available upon request). Oxy-Hb was primarily used in the analysis, because it is said to have a higher hemodynamic response than other signals (Huppert et al., 2006). First, each channel was pre-processed for oxy-Hb time series data. Channels with signal fluctuations of 10% or less were considered to have poor measurement quality and were excluded from the analysis. After exclusion, wavelet minimum description length (Wavelet-MDL) was applied to the remaining channels (Jang et al., 2009). Wavelet-MDL is a method for removing global trends due to respiration, heartbeat, and vascular motion, and does not require the definition of Hz for optional drifts.

Basis functions used for GLM analysis were generated from the hemodynamic response functions (HRFs) h(τ_p, t) which were created as in Equation 1 (Friston et al., 1998).

where t represents a point in the time series, τ_p represents the first peak delay, and τ_d represents the second peak delay. A is the amplitude ratio between the first and second peaks and was set to 6 as in typical fMRI studies. τ_p was adjusted as described below according to the adaptive GLM (Uga et al., 2014), and τ_d was set to 10 s after τ_p as in typical fMRI studies. Basis functions f(τ_p, t) were generated by convolving the HRF h(τ_p, t) with a boxcar function u(t),

where the symbol ⊗ denotes convolution integral. The basis functions were used to compose each regressor as described below.

For each channel, GLM analysis with regression to the basis functions was performed after preprocessing. The GLM analysis is generally presented in Equation 3,

where Y is a vector with M elements representing the hemodynamic response signal intensity corresponding to the oxy-Hb parameters. X is an M×N design matrix with N being the number of regressors. β is a coefficients vector with N elements. In the current study, β₁ corresponds to the regressor for product image presentation, β₂, for typing WTP, and β₃ represents a constant.

Finally, the least-squares estimation of β is given in Equation 4.

where X^T denotes the transpose matrix of X.

To apply the personalized adaptive GLM, which will be described later, t-values (t_value) were obtained to determine the optimized τ_p for each participant so that t_value reached maximum. t_value was calculated from the regression coefficient β and the residual ε using Equation 5,

where c is the contrast vector ([1, 0, 0] in this case) and determines the components of the regression coefficient β. Note that the thus-obtained t_value was not used for null hypothesis testing but as an indicator for τ_p optimization.

The HRF used to model changes in the blood oxygenation level dependent (BOLD) signal in fMRI studies is a regression function that emulates the actual hemodynamic response being measured. Nowadays, HRF is also often used in fNIRS. Boynton et al. (1996) showed that a gamma function with two free parameters, τ_p and τ_d, adequately represents hemodynamic responses.

Uga et al. (2014) reported an adaptive GLM that yields the most effective HRF by selecting the optimal τ_p. It has been reported that the WTP evaluation time: reaction times (RTs) were influenced by the WTP from individual to individual (Dezwaef et al., 2020). It can be predicted that different RTs will affect τ_p. Thus, we compared the R² values for conventional and personalized adaptive GLM in this study and found that the overall median R² was 0.323 for GLM with a fixed τ_p of 6, 0.800 for the personalized adaptive GLM of the DIY condition and 0.821 for a personalized adaptive GLM of the Non-DIY condition. Thus, in this study, we optimized HRF to each individual (personalized adaptive GLM), the details of which are to be described in “Preparation for GLM analysis” section as we mentioned above. This method identifies HRF to represent brain functions by selecting the optimal τ_p for each individual.

In this study, we have six regressors corresponding to the six products. The optimum τ_p for each participant was calculated for the hemodynamic response data of those six products. The value of τ_p varied between individuals from 2 to 21 s, which includes product image display and WTP typing. To reduce complication, the second peak delay τ_d and amplitude ratio A were set to the typical values (τ_d = 10 s, A = 6). The appropriate τ_p was specified based on t-values. The resulting τ_p for each individual is shown in Figure 3.

Then, hemodynamic response data for the six products were blocked into two conditions (DIY, Non-DIY). Brain activation was obtained as β values by applying the value of τ_p calculated earlier to the blocked data (Plichta et al., 2006, 2007). Figure 4 shows the observed hemodynamic response and the predicted response, which was created by combining regressors adjusted for each participant with the personalized adaptive GLM. Thus, four β values for two conditions (DIY and Non-DIY) and two constants were calculated. β₁ (DIY condition) and β₄ (Non-DIY condition) were subjected to second-level analysis.

For second-level analysis, we identified brain regions associated with the IKEA effect. Specifically, two β values for two conditions (DIY and Non-DIY) were subjected to second-level analysis. One-sample t-tests against zero were performed across participants for the null hypothesis that there is no change of hemodynamic response in a given channel. We used IBM SPSS statistical version 25 (IBM Inc., Armonk, NY, USA) and MATLAB2021 (The Mathworks Inc., Natick, MA, USA) to analyze behavioral and hemodynamic response data.

The WTP scores typed in the WTP evaluation task were subjected to a paired t-test for both conditions. This was to confirm that the IKEA effect occurred in all participants who subsequently underwent fNIRS data analysis.

In the second-level analysis-1, one-sample t-tests against zero (two tails) were conducted for both DIY and Non-DIY conditions in order to visualize the cortical activation under both conditions. In this analysis, we confirmed the putatively activated channels based on the effect sizes. It was necessary to consider a reasonable effect size for the data of 24 participants, to that end G^*Power (release 3.1.9.7) was used (Faul et al., 2007). Power analysis was conducted under the conditions of sample size = 24 and one-sample t-test with α = 0.05 and power = 0.8 and a reasonable effect size of 0.60 was obtained based on the study by Cohen (1992). For this reason, we defined a reasonable effect size of 0.60 or greater as activation of brain function in this analysis. Traditionally, p-values are used as an index in functional brain analysis. However, p-values are an index for judging the rejection of the null hypothesis but not to judge the acceptance of the alternative hypothesis (Cohen, 1994). This is the first study to measure the IKEA effect in terms of cortical activation. Considering the exploratory nature of our study, we wanted to reduce beta (Type II errors) as much as possible. Therefore, we decided that it was justifiable to conduct the analysis using effect sizes in this study.

Subsequently, in the second-level analysis-2, we compared the cortical activation between DIY and Non-DIY conditions. The channels with an effect size of 0.60 or larger in the one-sample t-test in either of the two conditions were defined as Channels of Interest, and a paired t-test was conducted. As with second-level analysis-1, an effect size of 0.60 or larger was defined as activation.

We performed a paired t-test on the WTP scores between both conditions. The results indicated significant differences between conditions as shown in Figure 5 [t₍₂₃₎ = 3.018, p = 0.007, d = 0.64]. Thus, we confirmed that the IKEA effect occurred for 24 participants.

We performed one-sample t-tests against zero (two tails) to examine brain activation in each condition as second-level analysis-1 (Figure 6A). Channel 46 was activated in both DIY and Non-DIY conditions: x = 40, y = 64, z = −3 in MNI coordinates; R-IFG in LBPA40; and right FPC in BA [DIY condition: t₍₂₃₎ = 3.11, p < 0.01, d = 0.65; Non-DIY condition: t₍₂₃₎ = 2.94, p < 0.01, d = 0.61]. Channels activated only in the DIY condition were Channel 39: x = −48, y = 49, z = 10 in MNI coordinates; L-IFG in LBPA40; and L-DLPFC in BA [t₍₂₃₎ = 2.95, p < 0.01, d = 0.62] and Channel 49: x = −38, y = 63, z = −2 in MNI coordinates; left middle frontal gyrus (L-MFG) in LBPA40; and left FPC in MRIcro [t₍₂₃₎ = 3.77, p < 0.01, d = 0.79].

Subsequently, paired t-tests were performed to compare brain activation between the two conditions as second-level analysis-2 (Figure 6B). Channel 49 was activated more highly in the DIY condition than in the Non-DIY condition [t₍₂₃₎ = 2.88, p < 0.01, d = 0.60].

The purpose of this study was to examine the cognitive mechanisms of the IKEA effect. The results confirmed the activation of the R-IFG, L-IFG, and L-MFG in the DIY condition and of the R-IFG in the Non-DIY condition. These results suggest that two regions, the L-IFG and L-MFG, are associated with the IKEA effect. These areas correspond to the FPC and L-DLPFC, respectively, when classified by the Brodmann area. Below, we discuss the brain regions found activated in the current study and consider the possible cognitive mechanism underlying the IKEA effect.

First, the R-IFG activations in both conditions were relevant to the consideration of the WTP evaluation for all products. The R-IFG has been shown to be associated with economic decision-making in multiple studies using fMRI (Plassmann et al., 2007; Knutson et al., 2008; Christopoulos et al., 2009; Hare et al., 2010). In addition, the current results are in line with a previous fMRI study showing R-IFG activation associated with the endowment effect (Votinov et al., 2010). Many previous studies have mentioned that the right PFC is significantly activated for WTP measurements. However, in the present study, the right PFC activation for the IKEA effect, found in the DIY minus Non-DIY contrast, was significant while the activation for WTP was not. Therefore, it is possible that the left PFC activation might reflect a composite cognitive reaction for WTP in combination with the IKEA effect.

Next, possible roles of the L-IFG and L-MFG, which were activated only for DIY products, will be discussed. The L-IFG has been associated with past memory recall (Clark et al., 2015). In this experiment, the recall of DIY memories was expected to occur while participants were viewing pictures of the products during the WTP evaluation. In this sense, the L-IFG activation is relevant and may reflect the memory recall function occurring during the WTP evaluation of DIY products. The L-MFG is said to be related to encoding schema on memory retrieval (Webb et al., 2016). In this sense, it is reasonable to infer that the L-MFG was activated to possibly recall the DIY experience and encode it into monetary values.

Moreover, the activations in the FPC and in the L-DLPFC in the DIY condition were similar to those from previous studies measuring brain functions related to attachment (Yargiçoglu, 2022). Considering the report by Atakan (2011) describing that attachment occurs through DIY activities, these activations suggest that the brain regions related to attachment were recruited for the DIY condition, which probably evoked attachment to the DIY products assembled by each participant. Based these results, we suggest that the attachment to the products were generated during the DIY experience and the memory of the DIY experience was recalled by looking at pictures of the DIY products during the WTP measurement. On the other hand, Kikuchi et al. (2021) described a positive correlation between brain areas related to memory (PCC–anterior hippocampus connectivity) and evaluation of attachment and revealed that greater product attachment evoked increased connectivity in memory-related regions. Unfortunately, the current study could not examine activations of these areas because they are located beyond the reach of near-infrared light. Thus, we have to note that the current study presents only a partial view of the neural substrates underlying the IKEA effect that possibly involves an association between attachment and memory.

We must note that the IKEA effect is considered to be a compound effect that is a mixture of various cognitive biases. First, the endowment effect is considered to be similar to the IKEA effect (Kahneman et al., 1990). Norton et al. (2012), however, identified the difference between the two effects by recognizing that the IKEA effect leads to increased WTP scores more than the simple endowment effect does. Second, Shmueli et al. (2012) found that the IKEA effect increases product valuations depending on the difficulty of the assembly process. Third, Ling et al. (2020) showed that the IKEA effect is magnified by increasing consumer choice opportunities in product completion. Gibbs and Drolet (2003) found that consumers' energy levels increase the likelihood of selecting DIY products. In other words, the IKEA effect may increase with the participant's energy level. We believe that adjusting for these factors and conducting neuroimaging experiments may shed more light on the neural mechanism behind the IKEA effect.

Furthermore, the two factors, memory and attachment, which are considered important cognitive components of the IKEA effect, may also be related to other experiential consumption, such as the experience of eating. In the future, we believe that the present experimental design incorporating fNIRS can be used as a reference for examining neural substrates for other types of experiential consumption. To build on this, further research on quantifying the value of experiential consumption, taking advantage of the many strengths of fNIRS, is required to further investigate the mechanism of the IKEA effect. We believe that such research can be conducted from two promising perspectives. The first is to conduct research under conditions that more closely approximate daily life. The second is to conduct research using simultaneous measurements. In fact, in recent years, there has been increasing research on simultaneous measurement using fNIRS and other devices, as well as research on using fNIRS measurements under conditions similar to daily life (Pinti et al., 2015; Bhatt et al., 2019; Sargent et al., 2020). Building upon such research will make it possible to explore the mechanism of the IKEA effect and quantify the value of experiential consumption by taking advantage of these fNIRS strengths.