Automated Detection of Decision-Making Style, Based on Users’

Online Mouse Pointer Activity

Marcus Cheetham

1a,*

, Catia Cepeda

, Hugo Gamboa

, Christoph Hoelscher

and Seyed Abolfazl Valizadeh

3e,*

Department of Internal Medicine, University Hospital Zurich, Zurich, Switzerland

LIBPhys (Laboratory for Instrumentation, Biomedical Engineering and Radiation Physics), Faculdade de Ciências e

Tecnologia, Universidade Nova de Lisboa, Caparica, Portugal

Chair of Cognitive Science, ETH Zurich, Clausiusstrasse 59, 8092 Zürich, Switzerland

Contributed equally to the paper

Keywords: Decision-Making Style, Personality, Computational Recognition, Computer Mouse, Pointer, Machine

Learning.

Abstract: Decision-making (DM) and online activity go hand in hand in many domains of everyday life (e.g., consumer

behaviour, financial and investment choices, career development, health and psychological well-being). DM

style refers to consistent behavioural patterns in the way different individuals approach DM situations. In this

study, we explored the feasibility of inferring DM style from the trace of mouse cursor (or pointer) activity

that users generated while performing an online task (the task required no explicit DM). We focussed on

maximizing and satisficing DM style. Based on a set of spatial, temporal and spatial-temporal features that

were extracted from mouse activity data and on measures of DM style assessed with a conventional self-report

questionnaire, we modelled DM style in a supervised machine learning approach. The results show that the

models of DM style have between good and high predictive accuracy. Guided by these results, we propose

that this mouse-based method might play a useful role in computational recognition of DM style and merits

further development. Future work will test the ability of pointer-based models to meaningfully link

psychological measures of DM style to objective measures and outcomes of real-world DM situations.

1 INTRODUCTION

Decision-making (DM) style refers to consistent

patterns of behaviour in the way individuals approach

DM situations (Scott & Bruce, 2016). For example,

maximizers tend to approach DM situations by

expending a lot of effort in searching and processing

the choice alternatives in order to determine the best

choice (i.e., maximizing style). In contrast, satisficers

tend to search and process the choice alternatives in

order to determine the choice that is good enough

(i.e., satisficing style) (Schwartz, 2000).

Maximizing and satisficing style are considered

by some researchers as habit-based tendencies to react

in a certain way to specific DM situations (e.g., a

https://orcid.org/0000-0002-1055-3923

https://orcid.org/0000-0002-2998-976X

https://orcid.org/0000-0002-4022-7424

https://orcid.org/0000-0002-5536-6582

https://orcid.org/0000-0003-0856-8541

particular DM style when choosing an item from a

range of consumer goods on the internet but not in the

local grocery store) (Scott & Bruce, 2016). Instead,

recent data shows that maximizing and satisficing

tendencies may be general to DM situations across

different DM domains (i.e., consumer goods, health

and life decisions, finance) (Moyano-Diaz &

Mendoza-Llanos, 2021). Irrespective of whether DM

style is a situation-specific habit or a general

disposition (or trait) of personality (Thunholm, 2004),

and given that DM and online activity are inextricably

linked in many DM domains (Kokkoris, 2018), we

asked if maximizing and satisficing style can be

inferred from the unique trace of online activity

generated while users interact with digital technology.

Cheetham, M., Cepeda, C., Gamboa, H., Hoelscher, C. and Valizadeh, S.

Automated Detection of Decision-Making Style, Based on Users’ Online Mouse Pointer Activity.

DOI: 10.5220/0011716700003414

In Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023) - Volume 4: BIOSIGNALS, pages 281-286

ISBN: 978-989-758-631-6; ISSN: 2184-4305

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

281

Much of the work of developing the maximizing

and satisficing concepts and their various components

(e.g., experienced decision difficulty) has focussed on

their assessment (Cheek & Schwartz, 2016). The

main method of assessing DM style is the traditional

paper-and-pencil self-report questionnaire (Boyle,

2009). A drawback of this method is that it is

impracticable for assessing online users and

unscalable to the many different online DM situations

and domains. Depending on the area of application,

online DM style might be better served using an

automated procedure with the potential for near real-

time processing.

Research in computational personality

recognition (CPR) and biometrics shows that the

digital trace of a user’s hand-held computer mouse

can be used to automatically assess user personality

and identity (Ahmed, Awad, & Traore, 2007;

Meidenbauer, Niu, Choe, Stier, & Berman, 2022;

Zhao, Miao, & Cai, 2022).

Recent work shows also

that DM style can be inferred from various data

sources, such as bodily and facial behaviour

(Connors, Rende, & Colton, 2013; Guo, Liu, Wang,

Zhu, & Zhan, 2022). However, it is not clear whether

and how well the activity of a mouse cursor, or

pointer, could be used for predictive modelling of

maximizing and satisficing style.

Numerous features can be extracted from the

movements and clicks of the pointer activity (Cepeda,

Dias, Rindlisbacher, Gamboa, & Cheetham, 2021;

Cepeda et al., 2018; Gamboa & Fred, 2003).

These

include temporal features (e.g., acceleration), spatial

features (e.g., curvature of trajectory) and composite

features based on complex spatial-temporal mouse

movements (e.g., hovering patterns) (Cepeda et al.,

2021).

We explored the feasibility of using pointer

activity data to predict users’ maximizing and

satisficing style. To this end, we acquired mouse

activity data while users performed an online task and

automatically extracted a broad set of temporal,

spatial and composite features (Cepeda et al., 2021).

The DM style of users was measured by self-report

questionnaire (Turner, Rim, Betz, & Nygren, 2016).

We then modelled the feature and style data in a

supervised machine learning approach.

The resulting predictive models of DM style

demonstrated between good and high accuracy.

Guided by these initial models, we suggest that

further development of this pointer-based method

might contribute to computational recognition of DM

style. We consider the potential application of this

method and future work to develop it further.

2 METHODOLOGY

2.1 Participants

N=79 (mean age=23.8, SD=4.07; 58 female) healthy

individuals, native speakers of Standard German,

with normal or corrected-to-normal vision and no

reported neurological or psychiatric illness

participated. Each participant gave written informed

consent and received 20 Swiss Francs (or course

credits if a student) for participation. The local Ethics

Committee waived the study (KEK Nr.: 2022-00713).

2.2 Dataset

Pointer features were extracted from data acquired

while participants engaged in the online task of

completing a digital German-language version of the

34-item Maximising Inventory (MI) (Turner et al.,

2016). The MI has three scales: decision difficulty,

alternative search, and satisficing, the first two of

which capture two different components of

maximizing behaviour. Participants rated each item,

using a 5-point Likert-type scale (1= “strongly

disagree” to 5= “strongly agree”). Cronbach's alpha

for each scale was 0.71, 0.87, 0.73, respectively

(comparable to the original English version).

2.3 Procedure

All participants were tested individually in a small,

quiet and dimly lit experimental room. The

experiment lasted approximately 60 min. First,

informed consent and demographic data was

collected. Then a 1 min. resting baseline was

performed at the beginning of the experiment to

facilitate laboratory adaptation. The questionnaire

was administered using LimeSurvey, an open-source

survey web app (LimeSurvey). After completion of

the survey, participants were informed that all data,

including mouse data, would be analysed and that

they could withdraw their data from the study if they

wished without stating a reason for doing so.

2.4 Data Acquisition

A web browser extension acquired the mouse activity

(Cepeda et al., 2019). The data included the (x,y)

coordinates of mouse position in pixels, the

questionnaire item and mouse events (click or no

click) associated with each mouse position,

timestamp (ms.), and click duration (ms. between

button press and release). Data were stored in the

MongoDB database (MongoDB) before exporting for

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

282

pre-processing. Features were extracted from the

mouse data and checking data consistency (within

and across participant data), data merging and

checking for missing data followed an automated

procedure (Cepeda et al., 2021).

2.5 Data Analyses

We used non-linear Random Forest (RF) regression

(Ho, 1998), with 108 temporal, spatial and composite

features as independent variables and the scores of

each scales as dependent variables.

Model input data was normalized (Gopal Krisna

Patro & Sahu, 2015). We set the number of decision

trees to 1000 and ran the bagging procedure 100

times. A 50% train-test split procedure was applied

(i.e., a conservative approach to avoid overfitting)

(Kuhn & Kjell, 2018). RF performance is relatively

robust against parameter specifications (Probst,

Bischl, & Boulesteix, 2019). The train-test procedure

selects features that optimize model accuracy and

resist non-informative predictors (Kuhn & Kjell,

2018). The bagging procedure is used for training

decision trees (Efron, 1994). The individual bootstrap

trees were aggregated to compute a final prediction of

performance for each model (Hastie, 2009; Ho, 1998;

James, 2013). Using the bagging procedure, RF

achieves higher, more stable predictive accuracy with

limited risk of overfitting (Valizadeh, Hanggi,

Merillat, & Jancke, 2017; Valizadeh, Riener, &

Jancke, 2019).

We used Mean Absolute Percentage Error

(MAPE) (i.e., difference between the actual value and

predicted value divided by actual value) for model

evaluation. As a percentage value, MAPE simplifies

comparison of performance across models. To

present results, we converted MAPE by computing

100-MAPE so that higher values indicate better

performance (Kleinbaum, Kupper, Muller, & Nizam,

1998).

First, we computed prediction models for each

scale using all features in each model. Second, we

computed 12 models, one for each combination of

scale and feature type (temporal, spatial, composite,

as well as all features together). Third, we performed

Welch’s ANOVA (Delacre, Leys, Mora, & Lakens,

2019; Welch, 1938), with scale (3 levels; decision

difficulty, alternative search, satisficing) and feature

type (4 levels; temporal, spatial, composite, all

features) as independent factors, and MAPE as

dependent variable. We tested for differences in

model accuracy between the scales and between the

types of features. The alpha threshold was set to .001

We used Scikit toolbox in Python for RF testing

and training (Pedregosa et al.) and R Version 4.0.2 for

analyses (Welch’s ANOVA) and creating figures (R-

Core-Team, 2022).

3 RESULTS

The mean scores of the scales were M=2.96

(SD=0.53) for decision difficulty, M=3.56 (SD=0.70)

for alternative search, M=3.45 (SD=0.45) for

satisficing.

The prediction models for the scales, with all

features in each model, show accuracies of 81% for

alternative search, 85% for decision difficulty and

88% for satisficing (see Fig. 1 and Table 1).

Table 1: Summary of the prediction models for each

combination of feature type and scale.

Feature type Scale

Alternative

(

)

Decision

difficult

(

)

Satisficing

(

)

All features 0.81 0.85 0.88

Com

osite 0.82 0.84 0.88

Spatial 0.82 0.85 0.89

Tem

oral 0.82 0.84 0.88

Note: % indicates MAPE after transforming this by 100-MAPE so

that a higher value corresponds to smaller prediction error.

Figure 1: Violin plots showing the median, variability and

probability density of prediction models for the decision

difficulty, alternative search and satisficing scales. All the

features were used as input in each model.

The F-Welch tests showed a significant difference

in model accuracy between these scales, F

(2,779.35)

1164.82, p < .001., and that model performance

accuracy was not significantly different between

Automated Detection of Decision-Making Style, Based on Users’ Online Mouse Pointer Activity

283

feature types, F

(3, 663.18)

=1.17, p=0.32 (see Fig. 2

and 3).

Overall, the results suggest between good and

high performance of all prediction models (i.e., a low

degree of error between the predicted and actual

values (see Table 1).

Figure 2: Violin plots showing the distributional

characteristics of the prediction models for each type of

feature and all features together (x-axis) against model

accuracy (y-axis).

Figure 3: Grouped violin plots showing the distributional

characteristics of the prediction models for each

combination of scale and feature (x-axis) against prediction

accuracy (y-axis).

4 CONCLUSIONS

In this paper, we propose the use of a pointer-based

method for automatically inferring the DM style of

users from the digital trace of their online activity. As

a proof of concept, we focussed on the psychological

constructs of maximising and satisficing, as we

considered these relevant for DM in various domains

(e.g., consumer behaviour, financial and investment

choices); other DM styles that could be investigated

(Cosenza & Ciccarelli, 2019; Harren, 1979; Janis &

Mann, 1977; Leykin, 2010; Saled, 2017; Weerasekara

& Bhanugopan, 2022). We examined three particular

scales of maximising and satisficing, but other scales

could be considered (Cheek & Schwartz, 2016).

All the models showed good predictive accuracy,

with the decision difficulty and satisficing scales

nearing high performance (Lewis, 1982). There was a

significant difference in model accuracy between the

scales but not between the three types of features used

for modelling. Overall, these initial findings speak in

favour of developing and testing this method further.

As an initial proof-of-concept study, we did not

examine whether there is any decay in model

performance when porting these models to other

unrelated online tasks. For this, we note that the

online task did not require any explicit form of DM

(though it did require reporting of behaviour related

to DM situations). A further analysis indicated good

stability of all models, suggesting that our sample size

is adequate for this proof-of-concept study.

This mouse-based (or pointer-based) method has

advantages over traditional self-report methods.

Remote pointer-based data collection is low cost,

easy to implement, nonintrusive (i.e., does not

interrupt the natural flow of user activity) and easily

scaled up. It can deliver results in near real-time and

could be re-applied to touch screen data.

Future work will seek to understand the impact on

model performance of factors that influence users’

pointer activity, such as technical factors (e.g.,

different mouse devices), environmental factors (e.g.,

ambient noise), task-related factors (e.g., nature and

goal of the online task) and individual human factors

(e.g., age). This could pave the way to developing

more robust models, though careful consideration

must be given to the choice of DM style (or related

constructs) (Misuraca, 2018) and the psychometric

properties of the self-report measures of DM style.

This pointer-based method could also contribute to

developing a better understanding of whether DM

style is a habitual tendency that is specific only to

certain situations or a more general pattern of

behaviour across time and situations (i.e., a

personality trait). Whether a habit or trait, the ability

of models based on a pointer device (e.g., computer

mouse, trackpad or digital pen) to meaningfully link

psychological measures of DM style to objective

measures of real-world DM needs to be evaluated.

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

284

ACKNOWLEDGEMENTS

This work was funded by the Universidade Nova de

Lisboa, Caparica, Portugal, and the University

Hospital Zurich, Zurich, Switzerland.

REFERENCES

Ahmed, A., Awad, E., &, & Traore, I. (2007). A New

Biometric Technology Based on Mouse Dynamics. IEEE

Transactions on Dependable and Secure Computing,

4(3), 165-179. doi:10.1109/tdsc.2007.70207

Boyle, G., & Helmes, E. (2009). Methods of personality

assessment. In The Cambridge handbook of personality

psychology (126 ed., Vol. 110, pp. 110-126): Cambridge

University Press.

Cepeda, C., Dias, M. C., Rindlisbacher, D., Gamboa, H., &

Cheetham, M. (2021). Knowledge extraction from

pointer movements and its application to detect

uncertainty. Heliyon, 7(1), e05873.

doi:10.1016/j.heliyon.2020.e05873

Cepeda, C., Rodrigues, J., Dias, M. C., Oliveira, D.,

Rindlisbacher, D., Cheetham, M., & Gamboa, H.

(2018). Mouse Tracking Measures and Movement

Patterns with Application for Online Surveys. In CD-

MAKE 2018: Machine Learning and Knowledge

Extraction (pp. 28-42).

Cepeda, C., Tonet, R., Osorio, D., Noronha, S., Hugo, P.,

Cheetham, M., & Gamboa, H. (2019). Latent: A Flexible

Data Collection Tool to Research Human Behavior in the

Context of Web Navigation. IEEE Access, 7, 77659-

77673. doi:10.1109/access.2019.2916996

Cheek, N. N., & Schwartz, B. (2016). On the meaning and

measurement of maximization. Judg. Decis. Mak. (11),

126–146.

Connors, B. L., Rende, R., &, & Colton, T. J. (2013).

Predicting individual differences in decision-making

process from signature movement styles: an illustrative

study of leaders. Front Psychol, 4, 658.

doi:10.3389/fpsyg.2013.00658

Cosenza, M., & Ciccarelli, M., & Nigro, G. (2019).

Decision-Making Styles, Negative Affectivity, and

Cognitive Distortions in Adolescent Gambling. J

Gambl Stud, 35(2), 517-531. doi:10.1007/s10899-018-

9790-y

Delacre, M., Leys, C., Mora, Y. L., & Lakens, D. (2019).

Taking Parametric Assumptions Seriously: Arguments

for the Use of Welch’s F-test instead of the Classical F-

test in One-Way ANOVA. International Review of

Social Psychology, 32(1). doi:10.5334/irsp.198

Efron, B., & Tibshirani, R. J. (1994). An introduction to the

bootstrap. New York: Chapman and Hall/CRC.

Gamboa, H., & Fred, A. (2003). An Identity Authentication

System Based On Human Computer Interaction

Behaviour. Paper presented at the PRIS.

Gopal Krisna Patro, S., & , & Sahu, K. K. (2015).

Normalization: A Preprocessing Stage. IARJSET

(10.17148/IARJSET).

Guo, Y., Liu, X., Wang, X., Zhu, T., & Zhan, W. (2022).

Automatic Decision-Making Style Recognition Method

Using Kinect Technology. Front Psychol, 13, 751914.

doi:10.3389/fpsyg.2022.751914

Harren, V. A. (1979). A model of career decision making

for college students. Journal of Vocational

Behavior(14), 119–133.

Hastie, T., Tibshirani, R., & Friedman, J. (2009). The

elements of statistical learning: Data mining, inference,

and prediction (2 ed.): Springer-Verlag.

Ho, T. K. (1998). The Random Subspace Method for

Constructing Decision Forests. IEEE Transactions on

Pattern Analysis and Machine Intelligence, 8(20), 832–

844. doi:doi:10.1109/34.709601.

James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013).

An introduction to statistical learning: With

applications in R: Springer-Verlag.

Janis, I. L., & , & Mann, L. (1977). Decision making: A

psychological analysis of conflict, choice and

commitment. New York: The Free Press.

Kleinbaum, D., Kupper, L., Muller, K., & Nizam, A.

(1998). Applied Regression Analysis and Multivariable

Methods. Pacific Grove, CA: Duxbury.

Kokkoris, M. D. (2018). Maximizing Without Borders:

Evidence That Maximizing Transcends Decision

Domains. Front Psychol, 9, 2664. doi:10.3389/

fpsyg.2018.02664

Kuhn, M., &, & Kjell, J. (2018). Applied Predictive

Modeling. New York, NY: Springer.

Lewis, C. D. (1982). Industrial and business forecasting

methods: A practical guide to exponential smoothing

and curve fitting. London; Boston: Butterworth

Scientific.

Leykin, Y., & DeRubeis, R. J. (2010). Decision-making

styles and depressive symptomatology Development of

the Decision Styles Questionnaire. Judgment and

Decision Making, 5(7), 506–515.

LimeSurvey. LimeSurvey: An Open Source survey tool

Retrieved from http://www.limesurvey.org

Meidenbauer, K. L., Niu, T., Choe, K. W., Stier, A. J., &

Berman, M. G. (2022). Mouse movements reflect

personality traits and task attentiveness in online

experiments. J Pers. doi:10.1111/jopy.12736

Misuraca, R., & Fasolo, B. (2018). Maximizing versus

satisficing in the digital age: Disjoint scales and the case

for “construct consensus. Personality and Individual

Differences, (121), 152-160.

MongoDB. MongoDB 5.0 Documentation. Retrieved from

https://docs.mongodb.com/

Moyano-Diaz, E., & Mendoza-Llanos, R. (2021). Yes!

Maximizers Maximize Almost Everything: The

Decision-Making Style Is Consistent in Different

Decision Domains. Front Psychol, 12, 663064.

doi:10.3389/fpsyg.2021.663064

Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., ,

Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P.,

Weiss, R.,, Dubourg, V., Vanderplas, J., Passos, A.,

Automated Detection of Decision-Making Style, Based on Users’ Online Mouse Pointer Activity

285

Cournapeau, D., Brucher, M., , . . . Duchesnay, E.

Scikit-learn: Machine learning in Python. The Journal

of Machine Learning Research, 12, 2825-2830.

Retrieved from https://scikit-learn.org/

Probst, P., Bischl, B., & Boulesteix, A.-L. (2019).

Tunability Importance of hyperparameters of machine

learning algorithms. The Journal of Machine Learning

Research, 20(1), 1934–1965.

R-Core-Team. (2022). R: A language and environment for

statistical computing. Retrieved from https://www.R-

project.org/.

Saled, M., Alhosseini, S. E., & Slambolchi, A. (2017). A

Review of Consumer Decision-Making Styles: Existing

Styles and Proposed Additional Styles. (07), 33-44.

Schwartz, B. (2000). The tyranny of freedom. American

Psychologist, 55, 79-88.

Scott, S. G., & Bruce, R. A. (2016). Decision-Making Style:

The Development and Assessment of a New Measure.

Educational and Psychological Measurement, 55(5),

818-831. doi:10.1177/0013164495055005017

Thunholm, P. (2004). Decision-making style: habit, style or

both? Personality and Individual Differences, 36(4),

931-944. doi:10.1016/s0191-8869(03)00162-4

Turner, B. M., Rim, H. B., Betz, N. E., & Nygren, T. E.

(2016). The maximization inventory. Judgment and

Decision Making, 7(1), 48-60. doi:10.1037/t45865-000

Valizadeh, S. A., Hanggi, J., Merillat, S., & Jancke, L.

(2017). Age prediction on the basis of brain anatomical

measures. Hum Brain Mapp, 38(2), 997-1008.

doi:10.1002/hbm.23434

Valizadeh, S. A., Riener, R., Elmer, S., & Jancke, L. (2019).

Decrypting the electrophysiological individuality of the

human brain: Identification of individuals based on

resting-state EEG activity. Neuroimage, 197, 470-481.

doi:10.1016/j.neuroimage.2019.04.005

Weerasekara, S., & Bhanugopan, R. (2022). The impact of

entrepreneurs’ decision-making style on SMEs’

financial performance. Journal of Entrepreneurship in

Emerging Economies. doi:10.1108/jeee-03-2021-0099

Welch, B. L. (1938). The significance of the difference

between two means when the population variances are

unequal. Biometrika, 29, 350–362.

Zhao, Y., Miao, D., & Cai, Z. (2022). Reading Personality

Preferences From Motion Patterns in Computer Mouse

Operations. IEEE Transactions on Affective

Computing, 13(3), 1619-1636. doi:10.1109/taffc.

2020.3023296.

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

286