Analyzing Customer Engagement with Gamification Approach in the Banking Sector Using the Rasch Model

2.1 Overview of the Rasch analysis

The Rasch model, a mathematical framework, utilizes probability estimates to evaluate the reliability and precision of rating scales, offering a robust methodology for assessment [25]. Considering the nonlinearity of raw scores, it is not reasonable to assume that the difference between two consecutive raw scores represents a consistent interval. This is one of the reasons why the Rasch measurement technique is employed. Some of these concerns are addressed by Wright [40] in their discussion. In an assessment of social skills using the Likert scale, let's consider the raw scores obtained by three individuals. Jim obtains a raw score of 20, Sue obtains a raw score of 22, and Jen obtains a raw score of 32. Based on the coding of raw score points, where Strongly Agree is assigned a code of 4, Agree as 3, Disagree as 2, and Strongly Disagree as 1, we can deduce that Jen demonstrates superior levels of social skills in comparison to the other alternatives. This inference is based on the fact that Jen's raw score corresponds to the highest raw score category, indicating stronger social skills. Rasch analysis employs a logit scale to rank the difficulty of items and the abilities of respondents, where higher scores indicate higher levels of ability or difficulty, while lower scores indicate the opposite. Before performing a Rasch analysis, there are a few preliminary steps that must be taken into account, including: item size, person size, Wright Map, match statistics, Rasch reliability index, and rating scale analysis. To demonstrate the practical application of Rasch's analytical methodologies in school psychology, this study illustrates their implementation by creating and piloting a self-report rating scale [33]. A logit scale was used in Rasch's study to rate participant ability and item difficulty, with higher scores denoting greater ability or difficulty and vice versa [41].

2.2 Gamification

Gamification is the strategic integration of games into non-gaming contexts or activities, aimed at enhancing customer engagement and fostering long-term retention [42]. The majority of studies investigating the use of gamification in banking have come to the conclusion that it significantly improves system performance and increases consumer engagement. The use of gamification to a variety of industries, including banking, will provide outstanding outcomes [16, 17]. This is because there is a clear link between gamification and the desire to utilize mobile banking services, which is supported by their widespread use. By employing a gamification strategy and creating suitable financial services, designers have the opportunity to enhance the appeal of banking operations. This, in turn, can lead to increased customer satisfaction and engagement. By comprehending essential concepts and effectively implementing a balanced selection of gamification techniques, it becomes possible to attain widespread consumer acceptance and value [16]. In a study by Rodrigues et al. [17], they developed a conceptual model to examine the effects of gamification on bank websites, and they demonstrated that it significantly affects web page design elements, user friendliness, and intent to use e-banking, among other things. Rahi et al. [43] study offered a solution for gamified internet banking and noted that by using gamification to make internet banking websites enjoyable and profit from the reward system, users' intentions to adopt the system and recommend it to others will improve social networking. According to related studies, gamification significantly enhanced customer interactions, use intention, and interest in social connections. For predicting behavioral intent in gamified commercial apps and linked enterprises, the researchers put out a conceptual model [44].

2.3 Demographic information

Previous research employed gender and age when using demographic data for various people in gamification systems. Orji [45], says that gender influences behavior, and men and women view behavior modification tactics differently. Lastly, Orji proposed gender as a reliable strategy and recommended the adoption of a gender-specific approach for design purposes. In a correlated investigation, Orji et al. [46] demonstrated that the techniques they employed were more effective in persuading women.

Oyibo et al. [47] investigated the role that culture plays in the influence of age and gender on the social influence of persuasive technology and demonstrated that there are significant differences between women and men and between younger and older people in collectivist and individualist cultures. Additionally, they furnished recommendations for tailoring personalization strategies based on cultural factors such as age and gender. PBL (points, badges, and leaderboards) is an interesting concept. According to some academics who looked into elderly people's gaming experiences, the majority of the currently available gamification systems are designed for younger audiences, have little relevance to parents, and cause them great stress when used [48]. Additionally, Denden et al. [49] have demonstrated that as children get older, their preferences, play motives, and play styles change from being performance-focused to being resolution, choice, and enjoyment-focused. They advise designers to concentrate more on objectives that favor setting and emotional connection as people age.

2.4 Gamification elements in banking

Recent advancements in the use of gamification systems have enhanced the requirement for diversified understanding between many people in numerous domains [15, 50, 51] Gamification is frequently linked to game elements, business applications, information systems, and entertaining tools. Fun-focused gamification endeavors to facilitate customer self-realization and foster sustained utilization of the system over the long term.

Table 1. Gamification elements studied in research

The review found 21 different gamification elements (see Table 1). Social networks and social interactions inside a system, for example, may fall under more than one part; however, for the sake of this review, each aspect is categorized under a different heading.

Table 1 categorizes various gamification elements used in banking to engage customers, such as announcements for user interaction, points based on performance that can be exchanged, rewards for completing challenges, rankings, badges for achievements, scores to motivate improvement, task completion menus, feedback mechanisms, leaderboards, promotional hunts, timed, games, levels offering personalized rewards, sharing features, social interaction opportunities, penalties for negative behaviors, avatars for personalization, lotteries, virtual rewards, epic meanings to foster belonging, informational services, and random rewards. Each element is designed to enhance user experience, foster community, and incentivize participation in the banking system.

2.5 Customer engagement

Extensive scrutiny of the concept of engagement has been conducted across multiple disciplines, including marketing, management, information systems, and information technology management [64]. Information systems and information technology experts view user engagement as a part of a more significant sedentary flow or condition that denotes a person's level of involvement and enjoyment in an activity [65].

Engaging customers strengthens long-term ties with them, which boosts corporate competitiveness [66]. Bowden states participation is a "psychological process" involving emotional and cognitive components [67]. According to Vivek et al. [68], customer engagement is the level of involvement and interaction with a company's goods and initiatives that customers or organizations initiate. Businesses create environments to enhance user involvement to significantly increase their chances of success Field [69].

Previous studies have looked into how customers assess the features of Islamic banks; Ahmad et al. [69] found that customers have a more favorable opinion of the quality of service in Islamic banks than in conventional banks. Customer preferences were consistently reflected in the aspects of tangibility and empathy that were used to evaluate the quality of the services. According to Estiri et al. [70], cost, product offerings, and other service qualities and value propositions are important factors in customer satisfaction but do not always translate into loyalty. Echchabi and Nafiu Olaniyi [71] observed that, among other things, branch location and service quality influenced Malaysian clients' preferences for Islamic banking. In a similar vein, Amin et al. [72] connected client loyalty with the bank image. The role of digital technology in transformation of regional models of household financial behavior analyzed by Zhavoronok et al. [73] found that innovation processes play an important role in national economies, which is largely due to the process of digitalization of financial relations in all developed countries.

4.1 Result

4.1.1 Formulation of the scale

Of the 26 dimensions of aspects related to customer involvement with a gamification approach in the banking sector using a Likert scale at intervals of 1-5 it shows that the scale most chosen by respondents is 5, which means that respondents' ratings tend to give a very agreeing assessment of items that are given. The following Figure 2 presents the results of responses from 451 respondents to the Likert scale items used in this study.

2.png

Figure 2. Respondent items

Based on the analysis of the formulation of the scale from 451 respondents showed that item 3 (undecided/neutral) and item 5 (strongly agree) respectively 22% and 21%. In items 1 and 2 regarding customer involvement, respondents are more likely to choose scale 3 (undecided/neutral). This indicates that customers already understand games but are still confused about using them when set in mobile banking, so this can provide benefits for banks and users. Meaningful customer engagement can be achieved through easy and fun and interactive game features. On the other hand, customer involvement in mobile banking applications can strengthen relationships with customers [39].

4.1.2 Data collection

In this research, a quantitative methodology is adopted, utilizing non-experimental designs to explore diverse phenomena. The collection of measurement data is of paramount importance as it enables the identification of relationships between empirical observations and quantitative mathematical expressions. To ensure the acquisition of primary data, an online questionnaire was meticulously designed and subsequently distributed via the user-friendly Google Forms platform. The questionnaire was thoughtfully conducted in the Indonesian language (Bahasa Indonesia) to specifically cater to the targeted participants from Indonesia. Table 3 provides an overview of the respondents' demographic profiles, encompassing factors such as gender, age, education, occupation, frequency of mobile banking usage (on a weekly basis), and weekly hours spent on the internet, social networks, video games, and similar activities. The sample exhibits a nearly equal distribution of male and female respondents, with the majority falling within the age range of 10 to 24 years old.

Table 3. The profile of research respondents

4.1.3 Data analysis

Linacre [78] propose an extensive set of indicators based on the Rasch model that can be applied to both individuals and items. These indicators encompass a range of psychometric properties, including mean square outfit (MNSQ), Z-standardized outfit (ZSTD), and point measurement correlation (PT-Measure Corr.). The evaluation process of the model commences by examining the MNSQ outfit value, which ideally falls within the 0.5 to 1.5 range, indicating a good fit for measurement. If the MNSQ value deviates from this range, attention is then shifted to the corresponding outfit ZSTD value, which should ideally be between -1.9 and 1.9, indicating reasonable data predictability. Internal reliability consistency, reflecting the average correlation among the items in the instrument, is assessed using Cronbach's α coefficient. A value approaching 1 signifies a favorable level of internal measurement consistency.

The study data was efficiently arranged using Microsoft Excel software and examined utilizing Winstep software version 3.73. After establishing suitable measurement intervals and satisfying validity and reliability criteria based on the Rasch model, the data underwent binary logistic regression analysis employing SPSS software. This particular methodology was selected to address the dichotomous nature of the questions answered by the participants.

4.1.4 Instrument reliability test

Table 4 displays the statistical overview of the instrument, encompassing person reliability and item reliability. The consistent findings confirm the accuracy and reliability of the measurements. The analysis yields two separate outputs, specifically person-output and item-output. The person table evaluates the statistical alignment of respondents with the data, whereas the item table assesses the compatibility of the instrument's items. As per Linacre [78], the mean value represents the signal-to-noise ratio within the data, with the mean coefficient indicating the square root of the ratio between the true person variance and the error variance present in the data. Boone et al. [79] stated that the index of the mean person and the mean item is a very important addition to the evaluation of the function of the measuring instrument.

Table 4. Statistical results of person and item reliability instruments

Based on the information presented in Table 4, the average person measure is recorded as 1.92. An average value above 0 indicates that the individuals or respondents possess higher abilities relative to the difficulty level of the items within the instrument. The individual reliability index is 0.94, while for items, it reaches 0.95. The responses from the subjects demonstrate strong consistency, and the overall reliability of the items is considered excellent, following the guidelines provided in standard Table 1 for person and item dependability.

The optimal values for INFIT and OUTFIT MNSQ approach 1, signifying a good fit, while the desirable values for INFIT and OUTFIT ZSTD are those close to 0, indicating reasonable predictability. After conducting a comprehensive analysis of the person and item tables, it is evident that the mean values of INFIT and OUTFIT MNSQ, as well as INFIT and OUTFIT ZSTD, align with the desired range. A higher separation value is preferred because it identifies a broader spectrum of subjects (those who are able versus unable) and item groups (difficult versus easy). Table 1 indicates a Separation Person value of 4.03, indicating that the instrument items are highly sensitive (very good) in covering the entire range of respondent abilities. The Separation Item value is 5.67, suggesting considerable variability among respondents and excellent detection of instrument item performance. The findings of the person and item validity tests, along with the MNSQ and PT-measure values, are depicted in Table 5. All items have exceeded the critical values for MNSQ and PT-Measure, thus affirming their validity. Linacre [78] suggests that the statistical output's sensitivity to outliers facilitates identifying and resolving any issues related to the appropriateness or compatibility between the data and the model. According to Boone et al. [79] Outfit Means-square, Outfit z-standard and Point Measure correlation values are the criteria used to determine item suitability. It is advisable to get the item repaired or replaced if it does not satisfy the requirements. The Outfit Mean Square (MNSQ) value received is 0.5 MNSQ 1.5, which is the criterion used to determine whether an item is appropriate. Table 5 displays item statistics in summary.

Table 5. Summary of item statistics

Table 6. Rating scale category measurement results

Analysis of the discussion of results and Table 5. Summary of Statistical Items can provide insight into the psychometric properties and performance of gamified customer engagement instruments in the banking sector. The table shows that the INFIT Mean square (MNSQ) value for each instrument item lies between 0.5-1.5 as shown in column 6. This shows that the items are all appropriate and do not need to be replaced, so it can be concluded that there is no there is an obvious disturbance or impact to the model while performing the statistical analysis.

4.1.5 Development of an instrument rating scale

The instrument's rating scale will be evaluated in the following stage. analysis on the results of this measurement is used to verify whether the choice rating used confuses the response or not. To see whether the rating scale function is running well, it can be seen through the measurement results in Table 6.

Based on the information presented in Table 6, the average person measure is recorded as 1.92. An average value above 0 indicates that the individuals or respondents possess higher abilities relative to the difficulty level of the items within the instrument. The individual reliability index is 0.94, while for items, it reaches 0.95. The responses from the subjects demonstrate strong consistency, and the overall reliability of the items is considered excellent, following the guidelines provided in standard Table 1 for person and item dependability.

The optimal values for INFIT and OUTFIT MNSQ are those that approach 1, signifying a good fit, while the desirable values for INFIT and OUTFIT ZSTD are those close to 0, indicating reasonable predictability. After conducting a comprehensive analysis of the person and item tables, it is evident that the mean values of INFIT and OUTFIT MNSQ, as well as INFIT and OUTFIT ZSTD, align with the desired range. A higher separation value is preferred because it allows for the identification of a broader spectrum of subjects (those who are able versus unable) and item groups (difficult versus easy). Table 1 indicates a Separation Person value of 4.03, indicating that the instrument items are highly sensitive (very good) in covering the entire range of respondent abilities. The Separation Item value is 5.67, suggesting considerable variability among respondents and excellent detection of instrument item performance. The findings of the person and item validity tests, along with the MNSQ and PT-Measure values, are depicted in Table 6. All items have exceeded the critical values for MNSQ and PT-Measure, thus affirming their validity. Linacre (2012) suggests that the statistical output's sensitivity to outliers facilitates the identification and resolution of any issues related to the appropriateness or compatibility between the data and the model [79]. According to Boone, et al. (2014), Outfit Means-square, Outfit z-standard and Point Measure correlation values are the criteria used to see the level of item suitability [78]. The Outfit Mean Square (MNSQ) value received is 0.5 MNSQ 1.5, and this number is the criterion used to determine whether an item is appropriate. Table 5 displays item statistics in summary.

Analysis of the discussion of results and Table 5. Summary of Statistical Items can provide insight into the psychometric properties and performance of gamified customer engagement instruments in the banking sector. The table shows that the INFIT Mean square (MNSQ) value for each instrument item lies between 0.5-1.5 as shown in column 6. This shows that the items are all appropriate and do not need to be replaced, so it can be concluded that there is no there is an obvious disturbance or impact to the model while performing the statistical analysis.

4.1.6 Development of an instrument rating scale

The instrument's rating scale will be evaluated in the following stage. analysis on the results of this measurement is used to verify whether the choice rating used confuses the response or not. To see whether the rating scale function is running well, it can be seen through the measurement results in Table 6.

Table 6 presents the measurement outcomes for each category in the rating scale. Category 1 corresponds to a score of 1 and consists of 2037 observations, accounting for 9% of the sample. The average measure for this category is -2.26, with an expected measure of -2.59. The Infit MNSQ is 1.66, and the Outfit MNSQ is 1.81. No threshold measures are available for this category. Category 2 represents a score of 2 and encompasses 1825 observations, making up 8% of the sample. The average measure is -1.26, while the expected measure is -0.96. The Infit MNSQ is 0.76, and the Outfit MNSQ is 0.77. The threshold measure for this category is -2.20. Category 3, with a score of 3, comprises 4768 observations, accounting for 22% of the sample. Both the average measure and expected measure for this category are 0.29. The Infit MNSQ is 0.75, and the Outfit MNSQ is 0.89. The threshold measure is -1.30. Category 4 represents a score of 4 and includes 4534 observations, making up 21% of the sample. The average measure is 1.74, while the expected measure is 1.70. The Infit MNSQ is 0.77, and the Outfit MNSQ is 0.67. The threshold measure is 1.02. Lastly, Category 5 corresponds to the highest score of 5, with 8484 observations, accounting for 39% of the sample. Both the average measure and expected measure for this category are 3.55. The Infit MNSQ is 1.28, and the Outfit MNSQ is 1.25. The threshold measure is 2.47. These findings provide a summary of the distribution and measurement characteristics for each category of the rating scale employed in the study.

The scale categories in Table 5 column 5 in the Category measure section show that each category is going well because on a scale of 1-5 the Category measure value increases proportionally and sequentially from -3.52 to 3.72. The existence of this monotonic increase indicates that the measurement has been going well. It is apparent that responders can discriminate between the answer options from very improper to very appropriate because there are no similar values in these five selections. Not all studies directly get good rating measurement results like this. Several studies related to Rasch analysis show that there are disproportionate conditions or unbalanced conditions, sometimes even the measurement of the rating scale is not shown as in research [80]. This can also be seen based on the probability of a person's ability as shown in Figure 3. Figure 3 demonstrates that if a person's ability is low, there is a high likelihood that they will respond with a rating of 1, while their ability is high, there is a higher likelihood that they would respond with a rating of 5. Figure 3 below depicts an ideal Rasch analysis model that can be used to compare the instrument's applicability as a whole.

Within a section of Figure 4, the scores of the instrument items are compared to the difficulty level measurement of the items, denoted by the blue line. The figure depicts the pattern of tested instrument item scores, which generally adheres to the ideal model of Rasch analysis, represented by the red line. While not flawless, the chart demonstrates that the instrument closely approximates the ideal Rasch pattern. This alignment between the instrument item scores and the Rasch ideal model is an encouraging discovery, indicating a reasonable level of agreement between the instrument and the targeted construct or variable. The closer the alignment between the instrument item scores and the Rasch ideal model, the stronger the measurement properties of the instrument.

3.png

Figure 3. Probability of person's ability

4.png

Figure 4. Overall instrument suitability results

4.2 Discussion

Comparing these results with previous research conducted by Lailiyah et al. [81], in analyzing self-assessment instrument items it shows that there are response patterns that are too far from the ideal model line curve. Therefore, additional analysis must be done, namely by gradually deleting things from the analysis that do not fit until the findings show that all of the items are consistent with the model. In Kadaryanto's study, which focuses on the design of teacher professional development programs utilizing the Rasch model, the analysis indicates a lack of alignment between the Item Characteristic Curve (ICC) and the desired pattern represented by the Red line (ideal model). In simpler terms, there are outliers observed in the distribution of respondents' answers to the survey questions. Ideally, the measured values should closely resemble the ideal model line. This suggests that some respondents may have provided careless or thoughtless responses to the survey questions [82]. Zansen's research utilized the Rasch measure to examine the fairness of the listening part across gender subgroups of a nationwide high-stakes computerized English examination. The NUDIF research' findings reveal that 12 items have gender-based DIF. The difference in ICC is not always discernible when pupils are separated into high-and low-achieving boys and girls [82].

The validity of the instruments used in further research can be assessed by examining the log values of each item or person. Deviations in logit values indicate a potential problem with a particular individual or item, indicating a need for revision or replacement. Ideally, the logit or size of each person and item should be closer to zero, indicating a better fit to the construct being measured. Another instrument validity indicator is the standard error of measurement (S.E. Measurement). A low standard error of measurement (less than 0.5) indicates high accuracy for an item or person, meaning that the measurements are reliable and precise. A standard measurement error between 0.5 and 1 is considered acceptable or moderately accurate, while a value exceeding 1.0 indicates poor accuracy for the person or item being measured. In addition to the standard error of measurement, the Outfit Mean Square (OUTFIT MNSQ) value provides insight into the instrument's validity. The ideal range for the OUTFIT MNSQ value is typically between 0.5 and 1.5. Values in this range indicate a good fit between the instrument items and the underlying construct. Departing from this range could suggest non-compliance or an issue with measurement that requires attention. The OUTFIT ZSTD metric is an additional means of evaluating the instrument's reliability. Values falling within the -2 to +2 range are deemed favorable, signifying a satisfactory alignment between the observed and expected responses derived from the Rasch model. Values outside this range may indicate potential problems with the measurement pattern or responses that deviate significantly from the expected model. Finally, Point Size Correlation is crucial in assessing the instrument's validity. Ideally, the Point Size Correlation should fall within the range of 0.4 to 0.85 to demonstrate sound validity. This correlation coefficient represents the degree of connection between the item's score on the instrument and the underlying construct being measured. Stronger correlations indicate more robust associations and offer supporting evidence of the instrument's proficiency in effectively measuring the intended construct [81].

By considering these validity indicators, researchers can assess the quality and feasibility of the instruments used in the study. If the instrument exhibits favorable log values, low standard error of measurement, suitable clothing mean squared values, clothing ZSTD values within the desired range, and satisfactory point size correlation, it indicates good validity. Conversely, if these indicators fall outside the recommended ranges, further investigation and potential instrument modification may be required to increase their validity.

Overall, the discussion emphasizes the significance of assessing the reliability of the research tools by considering numerous elements and criteria. Validity assessment ensures that the instrument accurately measures the intended construct and provides reliable and meaningful data for research purposes.

This study's limitations include its reliance on quantitative methods without thoroughly investigating respondents' diverse backgrounds. It leaves a gap in empirical insights for banks targeting specific community segments for their marketing efforts.