PERSONIFIKASI: Jurnal Ilmu Psikologi Vol.
17 No.
May 2026 ISSN: 2087-7447 .
| ISSN: 2721-0626 .
https://journal.
id/personifikasi Explaining Consumer Inaction in Boycott Movements: A Psychological Framework of Passivity During Periods of Social Change Akhmad Saputra Syarif 1*.
Dwi Yan Nugraha2,3, and Amya Bunga Fathiyah4 Faculty of Psychology.
Universitas Indonesia.
Indonesia Faculty of Psychology.
Universitas Gadjah Mada.
Indonesia Faculty of Psychology.
Universitas Islam Indonesia.
Indonesia Faculty of Humanities.
Universitas Jakarta Internasional.
Indonesia *Email: 1akhmad.
saputra@ui.
ARTICLE INFO
Article history:
Received September 13, 2025
Revised May 07, 2026
Accepted May 08, 2026
ABSTRACT
Research on collective action has primarily focused on participation, while the mechanisms underlying non-participation remain less understood.
This study examines why individuals remain inactive in boycott movements despite holding negative attitudes toward targeted companies, focusing on animosity toward firms associated with Israel.
Using a quantitative design, data from 212 Indonesian participants were analyzed using partial least squares structural equation modeling (PLSAeSEM) and artificial neural networkAeneural network regression (ANNAeNNR).
The PLSAeSEM approach was employed to examine explanatory relationships leading to purchasing behavior (PB).
The results indicate that several hypothesized predictors do not exhibit significant direct effects on PB, while indirect effects are partially supported through mediating mechanisms.
To complement the explanatory model.
ANNAeNNR results identified out of sight as the most important predictor of PB, followed by boycott attitude and urge of freedom.
This divergence suggests that psychological mechanisms differ between explanatory and predictive frameworks, with ANN capturing potential nonlinear relationships.
Overall, the study highlights the value of integrating PLSAeSEM and ANNAeNNR to better understand boycottrelated inaction in a non-Western context.
Key words: boycott, collective action, non-participation Journal Homepage: https://journal.
id/personifikasi This is an open access article under the CC BY SA license https://creativecommons.
org/licenses/by-sa/4.
INTRODUCTION
We will not go down.
In the night, without a fight.
You can burn up our mosques and our homes and our schools.
But our spirit will never die.
We will not go In Gaza tonight - Michel Heart Research on collective action has extensively examined why individuals participate in social movements, with influential frameworksAimost prominently the Social Identity Model of Collective Action (SIMCA)Aiemphasizing group-based emotions, and perceived efficacy (Louis et al.
, 2020.
van Zomeren et al.
, 2.
DOI: 10.
21107/personifikasi.
More recent work has extended these models to account for digital and networked forms of environments reshape processes of collective engagement (Chen et al.
, 2021.
Li, 2025.
Mongo, 2.
Subsequent extensions, including EMSICA (Louis et al.
, 2.
and ESIM (Thomas et al.
Despite these advances, the decision not to participate remains theoretically From a decision-making standpoint, individuals must first choose between action and inaction (Wright et al.
, rendering inaction a fundamental yet https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
understudied component of collective action Recent work in behavioral decisionmaking further indicates that inaction is not merely passive but reflects systematic cognitive tendencies such as inaction bias and omission effects (Fillon et al.
, 2.
Scholars have likewise argued that nonparticipation is not merely the absence of action but reflects distinct psychological processes (Stekelenburg & Klandermans.
More recent work conceptualizes inaction as an active psychological stance shaped by cognitive and contextual factors (Diesburg & Wessel, 2021.
Elliot et al.
, 2.
As Verba et al.
put it, individuals may refrain from collective action because they do not want to, cannot, or are not mobilized.
Mobilization processesAiboth consensus mobilization and action mobilizationAifurther shape whether individuals engage in or withdraw from collective action (Klandermans.
Empirical work examining these mechanisms, however, remains scarce, particularly in the context of consumer-based collective action such as boycotts.
Within the boycott domain, prior research has identified several psychological barriers to Yuksel .
distinguishes three such factorsAiout of sight, urge of freedom, and counterargumentsAias the principal explanations of inaction.
Subsequent work in political consumerism corroborates the 2 of 22 presence of analogous barriers in contemporary boycott movements (Copeland & Boulianne.
Hino, 2.
While this qualitative evidence offers important insights, it does not permit formal testing of predictive relationships among these Moreover, the bulk of research on collective action and inaction has been Western.
Educated.
Industrialized.
Rich, and Democratic (WEIRD) contexts, constraining its generalizability to non-Western settings (Atari et al.
, 2025.
Henrich et al.
, 2.
Recent commentary continues to underscore this limitation and calls for broader cultural representation in psychological science (Kryu et al.
, 2025.
yuskyl et al.
, 2.
Recent geopolitical tensions involving Israel and Palestine have catalysed widespread consumer boycotts across multiple countries, including Indonesia, furnishing a particularly apt context for examining collective action and non-participation in real-world settings .
Amalia et al.
, 2025.
Anwar et al.
, 2025.
Banjaransari, 2025.
Evelyn & Sekarasih, 2025.
Faizi et al.
, 2025.
Hino, 2022.
Husaeni & Ayoob, 2025.
Kristiningsih et al.
, 2.
Within this setting, public responses extend beyond participation in boycott actions and encompass varying degrees of inaction even among individuals who hold negative attitudes toward targeted companies.
Figure 1.
Proposes a model (Source: Researcher, 2.
DOI: 10.
21107/personifikasi.
https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
The present study addresses these gaps by examining non-participation in boycott actions within the Indonesian context.
Drawing on collective action theory and the political consumerism literature, recent integrative approaches emphasize the interplay among emotional, cognitive, and identity-based processes in shaping consumer activism (Ate.
Narayanan & Singh, 2025.
Ulfah et al.
Building on this perspective, the present study proposes that animosity toward Israel predicts perceptions of firmsAo unacceptable behavior, which in turn shapes boycott attitudes, intentions, and purchasing behavior.
Crucially, this pathway is theorized to be mediated by three internal psychological barriersAiout of sight, urge of freedom, and counterarguments .
ee Figure .
By formally testing this model, the study extends the collective action literature by demonstrating that inaction reflects distinct and measurable psychological processes.
In so doing, it advances a more comprehensive understanding of consumer behavior in boycott
movements, particularly in non-Western
METHOD
Procedure and Participants Data were collected through an online After providing informed consent, participants completed a brief demographic questionnaire followed by eight measurement scales and received a debriefing statement upon 3 of 22 The final sample comprised 212 respondents (Mage = 23.
SD = 4.
predominantly male .
7%).
Muslim .
8%),
and holding a bachelorAos degree .
4%).
9% reported participating in boycott Full demographic breakdown is reported in Table 1.
Measures Unless otherwise specified, all multi-item scales used a 5-point Likert response format ranging from strongly disagree to strongly agree, and reliability coefficients (CronbachAos ) are reported alongside each scale.
Animosity Towards Israel Animosity toward Israel was measured using an adapted scale from (C.
Kim.
Yan, et , 2.
An example item is.
AuI feel angry about the actions taken by Israel against PalestineAy ( = 0.
FirmsAo Unacceptable Behavior Perceptions of firms' unacceptable behavior were measured using a scale adapted from Abdul-Talib et al.
, which taps participants' evaluations of companies affiliated with Israel.
A sample item reads, "Companies affiliated with Israel deserve to have their products boycotted" ( = 0.
Out of Sight Out of sight was operationalized using binary .
es/n.
single-item measures adapted from Hoffmann .
to capture proximity to boycott-related impacts.
A sample item reads.
AuIs the city you live in close to a company affiliated with Israel that is subject to boycott?Ay Table 1.
Characteristics of the survey respondents (N = .
Characteristic Age .
Ae 39 years old.
M = 23.
SD = 4.
Gender Male Female Education High school BachelorAos degree MasterAos degree Religion Islam Christian Catholic Buddhist Boycott Status Non-Boycott Boycott Source: Personal research data, 2025.
DOI: 10.
21107/personifikasi.
Frequency Percentage (%) https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
Urge of Freedom Urge of freedom was assessed using a scale adapted from Dillard & Shen .
A sample item reads, "Messages about boycott calls interfere with my freedom to choose" ( = Counterargument Counterargument was measured using a scale adapted from (Abdul-Talib et al.
, 2.
A sample item reads, "Boycotting will harm Indonesia's economy" ( = 0.
Boycott Attitude Boycott attitude was measured using a scale adapted from (C.
Kim.
Yan, et al.
, 2.
example item is.
AuIt is beneficial to boycott products from companies affiliated with IsraelAy ( = 0.
Boycott Intention Boycott intention was measured using a scale adapted from (C.
Kim.
Yan, et al.
, 2.
An example item is.
AuI plan to boycott products from companies affiliated with IsraelAy ( = Purchasing Behavior Purchasing behavior was assessed using a single-item measure adapted from (C.
Kim.
Yan, et al.
, 2.
Participants indicated changes in their purchasing behavior over the past year on a 10-point scale ranging from 1 .
ot at all change.
to 10 .
ery much change.
The item read: AuCompared to your average purchases before the boycott campaign began on October 7, 2023, how have your purchases changed over the past year?Ay Data Analysis Data analysis proceeded in several stages.
First, descriptive statistics summarized participants' demographic characteristics and main study variables, followed by Pearson correlation analyses to examine bivariate relationships (Dancey & Reidy, 2.
and preliminary bias checks in IBM SPSS Statistics To test the proposed hypotheses.
Partial Least Squares Structural Equation Modeling (PLSAeSEM) was employed for its suitability in predictive modeling, theory development with complex models (Hair et al.
, and robustness under non-normal data (Hair.
Risher, et al.
, 2.
, using SmartPLS 9 (Henseler et al.
, 2.
Following current methodological recommendations.
DOI: 10.
21107/personifikasi.
4 of 22 both structural and measurement models were evaluated through explanatory power (RA), construct cross-validated redundancy (QA), fit indices (SRMR, d_G, d_ULS.
RMS.
NFI), the goodness-of-fit index (GoF), and robustness assessments for linearity, endogeneity, and heterogeneity (Hair et al.
, 2021, 2.
To complement PLSAeSEM and enhance predictive performance, an Artificial Neural Network (ANN.
Orry et al.
, 2.
Ai specifically Neural Network Regression (NNR.
Kovas et al.
, 2.
Aiwas implemented within a broader machine learning framework (Ghosh, 2022.
Jacobucci et al.
, 2023.
Pargent et al.
, 2023.
Vylez, 2.
A multilayer perceptron model was specified with animosity toward Israel (ATI), firms' unacceptable behavior (FUB), out of sight (OS), urge for freedom (UF), counterargument (CA), boycott attitude (BA), and boycott intention (BI) as inputs, and purchasing behavior (PB) as the Latent variable scores from PLSAeSEM served as inputs to account for measurement error and indicator loadings.
Data were normalized prior to estimation, and performance was evaluated via RMSE.
MAE, and RA.
ANN analyses were conducted in R 0 (Team, 2.
within the Positron (Posit Software.
Integrating ANN-based NNR with PLSAeSEM developments emphasizing hybrid machine learning approaches to improve predictive accuracy and assess variable importance in complex behavioral models (Baban & Baban.
Further methodological details appear in the supplementary materials.
RESULT
Descriptive Statistics and Correlations Normality assessment was conducted using skewness values to ensure data quality (Appelbaum et al.
, 2.
, although the PLS approach remains robust to non-normal distributions (Hair.
Risher, et al.
, 2.
All variables fell within the acceptable range (-3 to .
, indicating no severe deviations from normality (Heidary et al.
, 2.
Pearson correlations (Table .
revealed expected patterns: ATI.
FUB.
BA.
BI, and PB were positively interrelated .
s = 0.
348 to 765, all ps < 0.
, whereas urge of freedom (UF) and counterargument (CA) were https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
negatively associated with all boycott-related variables .
s = -0.
367 to -0.
Out of sight (OS) showed only a weak negative correlation with UF .
= -0.
138, p = 0.
and was otherwise unrelated to study variables.
These patterns suggest that boycott attitudes and 5 of 22 intentions are tightly coupled with purchasing behavior, while perceived freedom and counter-argumentative Table 2.
Descriptive statistics and correlation between variables Variables ATI
FUB
ATI
501** 0.
380** -0.
451** 0.
514** 0.
537** 0.
FUB
367** -0.
520** 0.
561** 0.
581** 0.
634** -0.
548** -0.
541** -0.
746** -0.
724** -0.
765** 0.
Minimum Maximum Mean Skewness Kurtosis Note.
*p < 0.
**p < 0.
***p < 0.
ATI: animosity towards Israel.
FUB: firmsAo unacceptable behavior.
OS: out of sight.
UF: urge of freedom.
CA: counterargument.
BA: boycott attitude.
BI: boycott intention.
PB: purchasing Source: Personal research data, 2025.
Common Method Variance (CMV) As all constructs were measured using a single data collection method, common method variance (CMV) was evaluated to ensure that the results were not biased by the measurement approach (Chin et al.
, 2.
Harman's singlefactor test indicated that the first factor accounted for 42.
05% of the total variance, below the conventional 50% threshold (Comrey, 1.
Thus, common method bias is unlikely to be a substantive concern in this Measurement Model Evaluation The measurement model was evaluated via variance-based CFA (Table .
All loadings 40, composite reliability values 70, and AVE values exceeded 0.
demonstrating adequate reliability and convergent validity (Hair.
Black, et al.
, 2019.
Hair et al.
, 2.
Model fit was acceptable based on SRMR .
068 < 0.
Hu & Bentler, 1.
SRMR
was prioritized over RMS theta given the latterAos known sensitivity to sample size and complex models (Maydeu-Olivares et al.
Discriminant validity was supported by HTMT values below 0.
85 (Henseler et al.
and cross-loadings (Table S.
, confirming that constructs are empirically The measurement model thus demonstrates adequate reliability and validity for subsequent structural analysis.
Table 3.
Measurement model evaluation Indicator VIF Animosity towards Israel Animo1 Animo2 Animo3 FirmsAo unacceptable behavior Firms1 Firms2 Firms3 Out of sight Distance1 Distance2 Distance3 DOI: 10.
21107/personifikasi.
AVE
Overall model fit measurement SRMR d_ULS Rms theta https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
Indicator VIF
AVE
6 of 22 Overall model fit measurement SRMR d_ULS Rms theta Urge of freedom Freedom1 Freedom2 Freedom3 Freedom4 Counterargument Counter1 Counter2 Counter3 Counter4 Counter5 Counter6 Boycott attitude Attitude1 Attitude2 Attitude3 Attitude4 Boycott intention Intention3 Intention4 Purchasing behavior Behaviour Note.
*p < 0.
**p < 0.
***p < 0.
LF: loading factor.
VIF: variance inflation factor.
: cronbach's alpha.
Ac:
composite reliability.
AVE: average variance extracted.
Source: Personal research data, 2025.
Structural Model Evaluation Prior to hypothesis testing, structural-level collinearity was assessed.
All variance inflation factor (VIF) values fell below the threshold of 10, indicating no multicollinearity issues (Field, 2.
the structural model was therefore deemed appropriate for hypothesis Table 4.
Path coefficient for direct effect
95% Confidence Interval
VIF
Lower Upper ATI i FUB < 0.
FUB i OS FUB i UF < 0.
FUB i CA < 0.
FUB i BA < 0.
OS i BA UF i BA CA i BA < 0.
FUB i BI
H10
OS i BI
H11
UF i BI
H12
CA i BI < 0.
H13
BA i BI < 0.
H14
FUB i PB
H15
OS i PB
H16
UF i PB
H17
CA i PB
H18
BA i PB
H19
BI i PB Note.
*p < 0.
**p < 0.
***p < 0.
VIF: variance inflation factor.
f2: fAesquare.
ATI: animosity towards Israel.
FUB: firmsAo unacceptable behavior.
OS: out of sight.
UF: urge of freedom.
CA: counterargument.
BA: boycott attitude.
BI: boycott intention.
PB: purchasing behavior.
Source: Personal research data, 2025.
Hypothesis Path Results (Table .
show that animosity toward Israel (ATI) significantly and positively predicted firms' unacceptable behavior (FUB.
B = 0.
557, t = 8.
327, p < 0.
, with a large DOI: 10.
21107/personifikasi.
effect size .
A = 0.
However.
FUB did not significantly predict out of sight (OS.
B = 0.
140, t = 1.
046, p = 0.
FUB significantly and negatively influenAe https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
7 of 22 Table 5.
Path coefficient for indirect effect
95% Confidence Interval
Lower
Upper
H20
FUB i OS i BA
H21
FUB i OS i BI
H22
FUB i OS i PB
H23
FUB i UF i BA
H24
FUB i UF i BI
H25
FUB i UF i PB
H26
FUB i CA i BA < 0.
H27
FUB i CA i BI < 0.
H28
FUB i CA i PB Note.
*p < 0.
**p < 0.
***p < 0.
VIF: variance inflation factor.
v: upsilon .
ATI: animosity towards Israel.
FUB: firmsAo unacceptable behavior.
OS: out of sight.
UF: urge of freedom.
CA: counterargument.
BA: boycott attitude.
BI: boycott intention.
PB: purchasing behavior.
Source: Personal research data, 2025.
Hypothesis Path ced urge for freedom (UF.
B = -0.
406, t = 542, p < 0.
and counterargument (CA.
= -0.
586, t = 8.
490, p < 0.
, suggesting that higher perceived unacceptable behavior weakens psychological resistance factors.
Conversely.
FUB positively predicted boycott attitude (BA.
B = 0.
297, t = 3.
965, p < 0.
and boycott intention (BI.
B = 0.
173, t = 2.
p = 0.
, though its direct effect on purchasing behavior (PB) was non-significant (B = 0.
059, t = 0.
697, p = 0.
Among mediators.
CA exerted strong negaAe Figure 2.
Structural model (Source: AuthorsAo elaboration using SmartPLS) tive effects on BA (B = -0.
506, t = 6.
597, p < .
and BI (B = -0.
244, t = 3.
372, p < 0.
whereas UF showed a weaker negative effect on BA (B = -0.
117, t = 1.
845, p = 0.
and no significant effect on BI or PB.
OS did not significantly predict BA.
BI, or PB.
strongly predicted BI (B = 0.
512, t = 5.
872, p < .
, while neither BA nor BI significantly influenced PB.
DOI: 10.
21107/personifikasi.
Mediation analysis (Table .
revealed that counterargument significantly mediated the relationship between FUB and BA (B = 0.
t = 6.
904, p < 0.
and between FUB and BI (B = 0.
143, t = 3.
298, p < 0.
, indicating substantively meaningful indirect effects.
Mediation through OS and UF, by contrast, was generally weak or non-significant.
These findings position counterargument as the https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
central mechanism explaining nonparticipation in boycott behavior .
ee Figure .
Quality of Model Predictive The coefficient of determination (R.
indicates that firmsAo unacceptable behavior (FUB) exhibits moderate explanatory power (R2 = 0.
, whereas boycott attitude (BA) and boycott intention (BI) show substantial explanatory strength (R2 = 0.
640 and R2 = 645, respectivel.
In contrast, out of sight (OS) presents a very low explanatory level (R2 = 0.
In terms of predictive relevance.
Q2 values suggest that most constructs possess adequate predictive capability, with the exception of OS (Q2 = -0.
, indicating a lack of predictive relevance for this variable (Hair et , 2021.
Hair.
Risher, et al.
, 2.
PLSpredict (Table .
indicated moderate overall predictive performance, with most RMSE/MAE values than the linear-model 8 of 22 benchmark (Shmueli et al.
, 2.
Overall fit indices were acceptable (SRMR = 0.
GoF = 0.
Klesel et al.
, 2019.
Wetzels & Odekerken, 2.
, indicating adequate explanatory capacity .
ee Table S.
Robustness Check Robustness checks confirmed model Linearity tests revealed mostly linear structural relationships .
s > 0.
, with only a few quadratic exceptions (Figure 3.
Table S.
, nonlinearity (Hair.
Risher, et al.
, 2.
FIMIX-PLS analysis identified a two-segment solution as optimal based on information criteria (AIC.
BIC.
CAIC) and entropy (EN = 773 > .
, indicating well-separated segmentation (Hair.
Black, et al.
, 2019.
Matthews et al.
, 2.
and notable structural differences across segmentsAievidence of latent heterogeneity within the sample (Table S5.
Table .
Table 6.
Predictive power
Model PLS
Model LM
Different (PLS Ae LM) RMSE
MAE
Q2 pred
RMSE MAE
Q2 pred iRMSE iMAE iQ2 pred FirmsAo unacceptable behavior (RMSE = 0.
MAE = 0.
QA pred = 0.
Firms1 Firms2 Firms3 Out of sight (RMSE = 1.
MAE = 0.
QA pred = -0.
Distance1 Distance2 Distance3 Urge of freedom (RMSE = 0.
MAE = 0.
QA pred = 0.
Freedom1 Freedom2 Freedom3 Freedom4 Counterargument (RMSE = 0.
MAE = 0.
QA pred = 0.
Counter1 Counter2 Counter3 Counter4 Counter5 Counter6 Boycott attitude (RMSE = 0.
MAE = 0.
QA pred = 0.
Attitude1 Attitude2 Attitude3 Attitude4 Boycott intention (RMSE = 0.
MAE = 0.
QA pred = 0.
Intention3 Intention4 Purchasing behavior (RMSE = 0.
MAE = 0.
QA pred = 0.
Behaviour Note.
The values in parentheses represent the LV (Latent Variabl.
prediction summary for each endogenous construct.
RMSE: root mean squared error.
MAE: mean absolute error.
Q2 pred: Q2 predict.
Source: Personal research data, 2025.
Indicator DOI: 10.
21107/personifikasi.
https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
Multi-Group Analysis (MGA) Multi-group analysis (MGA) revealed meaningful differences across gender .
ale vs.
and boycott status .
oycott vs.
Table S6.
Figure .
Gender-based comparisons yielded limited statistically significant differences, though several paths approached significance.
The FUBIeCA path showed a non-significant yet notable difference (B = 0.
154, p = 0.
, tentatively suggesting that males may be more reactive in forming counterarguments toward perceived corporate The FUBIePB path likewise revealed only marginal variation between male and female responses (B = -0.
113, p = 0.
9 of 22 In contrast, the comparison based on boycott status demonstrates more pronounced and statistically significant differences.
The relationship between animosity towards Israel (ATI) and FUB is significantly stronger in the boycott group (B = -0.
268, p = 0.
indicating that individuals engaged in boycott behavior are more sensitive to perceived corporate misconduct.
A similar pattern is observed in the relationship between FUB and boycott attitude (BA), where the boycott group exhibits a stronger negative effect (B = -0.
p = 0.
, suggesting a greater tendency to develop boycott attitudes in response to unacceptable corporate practices.
Figure 3.
Assessment of nonlinearity effects in the structural model (Source: AuthorsAo elaboration using SmartPLS) Table 7.
Results of FIMIX-PLS calculation Parameters Relative seg.
size (%) CronbachAos Alpha
Composite Reliability AVE
HTMT
Path ATI i FUB FUB i OS FUB i UF FUB i CA FUB i BA Original sample FIMIX PLS group 1 Measurement model Structural model < 0.
< 0.
< 0.
< 0.
< 0.
< 0.
DOI: 10.
21107/personifikasi.
FIMIX PLS group 2 < 0.
< 0.
< 0.
< 0.
https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
10 of 22 Parameters OS i BA UF i BA CA i BA FUB i BI OS i BI UF i BI CA i BI BA i BI FUB i PB OS i PB UF i PB CA i PB BA i PB BI i PB Original sample FIMIX PLS group 1 FIMIX PLS group 2 < 0.
< 0.
< 0.
< 0.
< 0.
< 0.
< 0.
< 0.
Coefficient determination (R.
FUB Note.
*p < 0.
**p < 0.
***p < 0.
ATI: animosity towards Israel.
FUB: firmsAo unacceptable behavior.
OS: out of sight.
UF: urge of freedom.
CA: counterargument.
BA: boycott attitude.
BI: boycott intention.
PB: purchasing AVE: average variance extracted.
HTMT: heterotraitAemonotrait ratio of correlations.
R2: RAesquare.
: All indicators meet the criteria.
Ae: Not all indicators meet the criteria.
Source: Personal research data, 2025.
Furthermore, the relationship between BA and PB differs significantly across boycott status groups (B = -0.
601, p = 0.
indicating that boycott attitudes translate more strongly into actual purchasing behavior among boycott participants.
These findings underscore behavioral consistency within the boycott group, where attitudes more directly drive consumption decisions.
Additional insights emerge from the mediaAe Figure 4.
Multi-group structural model across gender and boycott status (Source: AuthorsAo elaboration using SmartPLS) DOI: 10.
21107/personifikasi.
https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
tion analysis.
The indirect effect of FUB on boycott intention (BI) through CA is significant in the male group (B = 0.
184, p = 0.
but not in the non-boycott group, suggesting that counter-arguments play a more prominent mediating role among male participants.
Overall, the MGA results indicate that boycott status represents a more substantial source of heterogeneity than gender.
Individuals in the boycott group exhibit stronger and more consistent responses to corporate misconduct, particularly in shaping attitudes, intentions, and purchasing behavior .
ee Figure .
Neural Network Regression (NNR) Performance of the artificial neural networkAeneural network regression (ANNAe NNR) models was evaluated on training and testing datasets across 24 architectures (Table 11 of 22 Model comparison relied on root mean squared error (RMSE), mean absolute error (MAE), and the coefficient of determination (RA), with primary emphasis on testing performance to ensure out-of-sample validity.
TrainingAetesting differentials .
RMSE, iRA, iMAE) were further examined to assess stability and potential overfitting.
Predictive performance varied substantially across configurations.
Model ANNAeNNR 1 .
Ae 3Ae1 architectur.
emerged as the best performer, yielding the lowest RMSE .
and MAE .
and the highest RA .
on the testing set.
It also exhibited the smallest trainingAetesting discrepancy .
RMSE = 0.
iRA = Oe0.
iMAE = 0.
, indicating superior accuracy, stronger generalization, and minimal overfitting relative to more complex Table 8.
Comparison model ANNAeNNR
Model
ANNAeNNR
Architecture Training Data Set
RMSE
MAE
Testing Data Set
RMSE
MAE
Different (Training Ae Testin.
iRMSE iR2 iMAE Single hidden layer models Model ANNAeNNR 1 7Ae3Ae1 Model ANNAeNNR 2 7Ae5Ae1 Model ANNAeNNR 3 7Ae8Ae1 Model ANNAeNNR 4 7Ae10Ae1 Model ANNAeNNR 5 7Ae15Ae1 Model ANNAeNNR 6 7Ae20Ae1 Two hidden layers models Model ANNAeNNR 7 7Ae5Ae3Ae1 Model ANNAeNNR 8 7Ae8Ae5Ae1 Model ANNAeNNR 9 7Ae10Ae5Ae1 Model ANNAeNNR 10 7Ae10Ae10Ae1 Model ANNAeNNR 11 7Ae15Ae10Ae1 Model ANNAeNNR 12 7Ae20Ae20Ae1 Three hidden layer models Model ANNAeNNR 13 7Ae5Ae3Ae2Ae1 Model ANNAeNNR 14 7Ae8Ae5Ae3Ae1 Model ANNAeNNR 15 7Ae10Ae8Ae5Ae1 Model ANNAeNNR 16 7Ae12Ae8Ae5Ae1 Model ANNAeNNR 17 7Ae15Ae10Ae5Ae1 Model ANNAeNNR 18 7Ae20Ae15Ae10Ae1 Four hidden layer models Model ANNAeNNR 19 7Ae5Ae4Ae3Ae2Ae1 Model ANNAeNNR 20 7Ae8Ae6Ae4Ae3Ae1 Model ANNAeNNR 21 7Ae10Ae8Ae6Ae4Ae1 Model ANNAeNNR 22 7Ae12Ae10Ae8Ae5Ae1 Model ANNAeNNR 23 7Ae15Ae12Ae10Ae8Ae1 Model ANNAeNNR 24 7Ae20Ae15Ae10Ae5Ae1 Note.
ANNAeNNR: artificial neural network-neural network regression.
RMSE: root mean squared error.
MAE: mean absolute error.
R2: RAe square .
oefficient of determinatio.
The ANNAeNNR architecture refers to the network topology used in the prediction model.
Source:
Personal research data, 2025.
Importantly, model performance did not improve with increasing network complexity.
Models with deeper architectures .
wo, three, and four hidden layer.
generally exhibited larger trainAetest discrepancies, higher prediction errors, and lower explanatory power DOI: 10.
21107/personifikasi.
on the testing dataset.
Several complex models, such as Model ANNAeNNR 7 .
Ae5Ae3Ae.
, showed substantial performance deterioration between training and testing datasets .
RMSE = -2.
, suggesting pronounced overfitting.
These findings indicate that more complex https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
ANNAeNNR structures tend to memorize generalization performance.
In contrast, simpler architectures demonstrated more stable and reliable predictive behavior .
ee Figure .
To further interpret the contribution of each predictor variable within the best-performing 12 of 22 ANNAeNNR model .
Ae3Ae1 architectur.
, variable importance analysis was conducted using the variable importance function implemented in the H2o deep learning framework, which estimates the relative contribution of each predictor to the modelAos predictive performance.
Figure 5.
Neural network structure (Source: AuthorsAo elaboration using R and Positro.
The variable importance analysis (Table .
shows that out of sight is the most influential predictor, accounting for 22.
400% of overall model importanceAisuggesting that perceived proximity or visibility of boycott-related targets substantially shapes purchase behavior.
Boycott attitude .
000%) and urge of .
underscoring the role of cognitive and intentional processes in consumer decisionmaking.
Boycott intention, firms' unacceptable behavior, and animosity toward Israel In contrast, counterargument exhibited the lowest contribution .
400%), reflecting a comparatively smaller role within the ANNAe NNR framework.
Overall, variables related to DOI: 10.
21107/personifikasi.
behavioral proximity and boycott attitudes contributed more strongly to predictive performance than other motivational variables.
These findings align with the superior performance of the parsimonious 7Ae3Ae1 architecture reported in Table 8, reinforcing the importance of model simplicity for better Figure 6 illustrates the importance of the best-performing ANNAeNNR model .
Ae3Ae1 architectur.
Panels A.
B, and C show the progression of RMSE.
MAE, and deviance across training epochs for both training and validation datasets.
Prediction errors decreased substantially during early training and gradually stabilized as epochs Training curves consistently yielded lower error values than validation curves, https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
indicating that the model captured meaningful data patterns while maintaining stable RMSE and MAE declined sharply during the initial epochs before reaching a plateau, signaling convergence of the optimization process.
Although a gap persisted between training and validation performance, the discrepancy remained 13 of 22 moderate relative to more complex architectures in Table 8, supporting the conclusion that the 7Ae3Ae1 architecture achieved better predictive stability with reduced overfitting.
The deviance curve likewise showed a continuous downward trend, further indicating improved model fit throughout training.
Table 9.
Variable importance of ANNAeNNR model in predicting purchase behavior Importance Relative Scaled Variables Percentage (%) Rank Importance Importance Out of sight (OS) 224 .
Boycott attitude (BA) 220 .
Urge of freedom (UF) 193 .
Boycott intention (BI) 119 .
FirmsAo unacceptable behavior (FUB) 087 .
Animosity towards Israel (ATI) 083 .
Counterargument (CA) 074 .
Note.
ANNAeNNR: artificial neural networkAeneural network regression.
Variable importance was computed using the H2o deep learning variable importance procedure based on the trained ANNAeNNR model .
rchitecture 7Ae3Ae.
, which was identified as the best-performing model in Table 8.
Relative importance reflects the model-derived contribution of each predictor, whereas scaled importance represents normalized relative contributions.
Percentage values indicate each variableAos proportional contribution to the modelAos predictive performance in purchase behavior.
Source: Personal research data, 2025.
Figure 6.
Model validation and variable importance of the ANNAeNNR model architecture 7Ae3Ae1 (Source:
AuthorsAo elaboration using R and Positro.
DOI: 10.
21107/personifikasi.
https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
The ANNAeNNR model was estimated using the Tanh activation function with 1000 epochs and one hidden layer consisting of three The model additionally employed l1 regularization .
, l2 regularization .
, adaptive learning rate optimization, 3-fold cross-validation.
Variable importance analysis (Panel D) revealed that out of sight (OS), boycott attitude (BA), and urge of freedom (UF) contributed most strongly to the prediction of purchase behavior, whereas counterargument (CA) demonstrated the smallest contribution to the model.
DISCUSSION
The findings of this study extend prior work on collective-action inaction by integrating mechanisms within a single structural Consistent with Yuksel .
, the results demonstrate that multiple psychological factors simultaneously shape individuals' reluctance or willingness to engage in boycott First, animosity toward Israel significantly predicted perceptions of firms' unacceptable This finding aligns with the classical spillover logic articulated by Friedman .
, wherein consumers redirect anger toward associated entities when direct punishment of the primary target is infeasible.
Klein et al.
likewise argue that firms function as symbolic representatives of the target country, particularly when prior affiliations are salient.
The Indonesian context reinforces this mechanism: geopolitical conflict translates into marketplace responses.
Recent studies on consumer animosity and boycott behavior further confirm that such responses are contextually situated and shaped by the prevailing political-moral environment (Babu et al.
, 2025.
Bryckerhoff & Qassoum, 2021.
Davlembayeva et al.
, 2024.
Hino, 2022.
Kristiningsih et al.
, 2.
, underscoring that boycott behavior is embedded in lived sociopolitical conflict rather than reducible to individual preference.
Second, firms' unacceptable behavior exerted both direct and indirect effects on boycott attitude and intention.
The direct positive effects corroborate prior findings .
Abdul-Talib et al.
, 2.
showing that perceived corporate wrongdoing intensifies collective-action tendencies.
The present study DOI: 10.
21107/personifikasi.
14 of 22 extends this work by demonstrating that the relationship is contingent on mediating mechanismsAia pattern consistent with research on the ambiguous effectiveness of political brand communication, where corporate stances can be interpreted inconsistently depending on consumers' ideological alignment and contextual cues .
Flight & Allaway, 2024.
Guyvremont.
Jungblut & Johnen, 2022.
Ketron et al.
Kim & He, 2.
The role of counterargument aligns with bystander theory and collective-inaction frameworks (Klein et al.
, 2.
, which suggest that individuals weigh potential costs and consequences before engaging in collective When boycott actions are perceived as ineffectiveAior even harmful .
, economic consequences for local worker.
Aiindividuals generate counterarguments that diminish participation willingness.
This dovetails with recent behavioral decision-making research showing that perceived inefficacy and personal cost reduce engagement in prosocial or political consumption .
Lee, 2026.
Starke & Kelm, 2.
Recent literature on digital activism further suggests that individuals increasingly experience activism fatigue and cognitive overload in online political environments, strengthening counterargument formation and behavioral disengagement (Lane et al.
, 2025.
Simiti, 2024.
Valli & Nai, 2.
The welldocumented slacktivismAereal-action likewise helps explain why online moral agreement does not always translate into offline boycott behavior (C.
Kim.
Kim, et al.
Kim & He, 2.
Work on systemjustifying tendencies in political consumption further suggests that individuals may rationalize non-participation by minimizing personal responsibility or the perceived efficacy of collective action (Nakagoshi & Inamasu, 2.
By contrast, the urge of freedom can be interpreted through Psychological Reactance Theory (Brehm & Brehm, 1.
When individuals perceive social pressure to participate in boycotts, they may experience reactance and resist compliance in order to restore autonomy.
As Yuksel .
observes, boycott calls can paradoxically trigger resistance when framed as social obligation rather than autonomous choice.
https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
Recent empirical research extends reactance theory into digital environments, showing that persuasion resistance emerges on social media when users perceive moral pressure within online activism campaigns (Carpenter et al.
, 2026.
Yang & Kruschke.
Hashtag-activism backlash has likewise been documented when individuals reject symbolic participation demands perceived as socially coercive (Hong & Kim, 2021.
Wang & Zhou, 2.
The construct of 'forced-activism perception' further highlights how normative pressure online can reduce behavioral compliance and amplify psychological resistance (Lu & Liang, 2024.
Yang & Kruschke, 2.
Interestingly, out of sight did not significantly mediate the relationship between firms' unacceptable behavior and boycott outcomes, suggesting that perceived distance from boycott consequences is not decisive in the present context.
One plausible explanation lies in shared social identity: according to Social Identity Theory (Tajfel & Turner, 2.
, individuals may act on the basis of group affiliation rather than personal proximity to Recent extensions of identity theory emphasize moral identity in activism, with individuals engaging in collective action through internalized moral self-concepts rather than situational proximity (Leal et al.
, 2025.
Shultziner, 2.
Research on collective identity in digital movements further shows that online communities strengthen perceived group belonging, sustaining engagement even when personal costs are diffuse or unclear (Brown et al.
, 2024.
Greijdanus et al.
, 2.
In addition, the concept of global solidarity consumptionAiparticularly sensitive contexts such as Palestine-related activismAisuggests that individuals engage in boycott behavior as a form of transnational moral alignment rather than direct experiential proximity (Zejjari & Benhayoun, 2.
This finding adds nuance to prior work emphasizing psychological distance (Ram et al.
, 2024.
Trope & Liberman, 2.
, indicating possible boundary conditions in digitally mediated contexts of moral solidarity.
Furthermore, none of the variables significantly predicted purchasing behavior.
This pattern supports the well-established attitudeAebehavior gap articulated by the Theory DOI: 10.
21107/personifikasi.
15 of 22 of Reasoned Action (Ajzen & Fishbein, 1.
, which posits that attitudes and intentions do not always translate into actual behavior.
Within this framework, attitudes are determined by beliefs about the consequences of behaviorAi particularly whether those consequences are positive or negativeAisuch that individuals display more favorable attitudes when they expect positive outcomes (Mubin & Setyaningsih, 2.
Recent work in ethical consumption further highlights that the intentionAebehavior gap persists in digital-era activism, where moral intentions expressed online are often decisions(Azzopardi & Sluis, 2024.
Casais & Faria, 2.
In addition, research on boycott frequently experience cognitive dissonance between ideological support and habitual consumption patterns (J.
Kim et al.
, 2025.
Roth et al.
, 2.
Taken together, these findings suggest that structural constraints, habit formation, and competing preferences remain key barriers to translating boycott intentions into actual behavior.
Finally, the multi-group analysis revealed meaningful differences across gender and boycott status.
These findings align with prior research indicating that demographic factors shape collective-action tendencies .
Besta et al.
, 2024.
Nelson et al.
, 2.
More recent work further shows that gendered patterns of political consumption shape moral decisionmaking in boycott contexts, with women tending to display higher ethical sensitivity and a stronger orientation toward collective responsibility (Jansesberger & Lefkofridi.
Vasquez et al.
, 2.
More importantly, differences between boycott and non-boycott groups suggest that prior commitment shapes how individuals interpret corporate behavior and translate it into This finding underscores the importance of segmentation for understanding consumer activism, particularly in politically charged consumption environments (Ate.
Garg & Saluja, 2.
Overall, this study contributes to the literature by demonstrating that collective action in the context of boycotts is not solely driven by moral outrage but is simultaneously shaped by cognitive evaluation, perceived autonomy, and social identity processes.
The https://journal.
id/personifikasi PERSONIFIKASI: Jurnal Ilmu Psikologi 2026 (Ma.
, vol.
17, no.
Syarif, et al.
integration of these mechanisms yields a more comprehensive explanation of why individuals choose to engage or disengage from collective These findings suggest that boycott inaction should be understood not as a mere lack of motivation but as an active psychological process shaped by competing cognitive justifications, perceived autonomy threats, and digitally mediated social identities within contemporary political-consumerism CONCLUSION This study examined why individuals remain inactive in boycott movements despite holding negative attitudes toward companies associated with Israel, with particular attention to the role of animosity in shaping nonparticipation.
The findings indicate that animosity toward targeted companies is indirectly linked to boycott attitudes and intentions through distinct psychological Of the three motivational factors proposed by Yuksel .
, only urge for freedom and counterargument significantly mediated the relationship between perceived firm misconduct and both boycott attitude and intention, whereas out of sight showed no significant effect.
These results suggest that boycott inaction is not a passive response, but an active process shaped by cognitive and motivational dynamics, particularly those tied to perceived autonomy and argumentative The divergence from prior findings highlights the contextual specificity of inaction in politically sensitive boycott settings.
Future research should examine how animosityrelated cognition interacts with motivational mechanisms across diverse sociopolitical contexts to better understand when negative attitudes translate into active boycott behavior.
REFERENCES