According to Johnson and Wichern [8] in the hierarchical method of forming groups, there are two approaches, namely agglomerative hierarchical methods (Agglomerative Hierarchical Methods) and divisive hierarchical methods (Device Hierarchical Methods). The agglomerative method starts by assuming that each object is a cluster. Then the two objects that have the closest distance are combined into one cluster. The process continues so that in the end it will form a cluster consisting of all objects [9].

Agglomerative methods start with individual objects. Initially, there are as many clusters as objects. The most similar objects are grouped and then these initial groups are combined according to their similarities [10]. When similarity decreases, all subgroups are combined into one group. While divisive hierarchical clustering reverses the process of agglomerative hierarchical clustering, by starting with all objects in one cluster, and successively dividing each cluster into smaller ones [11]. The method that is often used is the agglomerative hierarchical method. The agglomerative hierarchical method algorithms used in group formation include:

In Ward clustering method, at every step, every possible union of cluster is considered, and those two clusters are combined together whose merger results in minimum increase in information loss. The Ward method is a method of forming clusters based on the loss of information due to merging objects into clusters. (1)SSE=∑j=1p(∑i=1nεij2-1n(∑i=1nεij2)2) where:

ε_ij: value for the i-th object on Cluster j

p: number of variables

n: the number of respondents in Cluster formed

Steps to solve the Ward method:

In the Complete linkage method, the distance between clusters is determined by the furthest distance between two objects in different clusters [8]. First start with all the points as individual clusters then merge two points that have the minimum distance between them. (2)d(ij)k=max(dik,djk) where:

d(ij)k: distances between sub-samples (ij) and Cluster k

sis: distance of sub-sample i and cluster k

djk: distance of sub-sample j and cluster k

In the Average linkage method, the distance between two Clusters is considered as the average distance between all members in one Cluster and all other Cluster members. The distance formula can be written in equation (3). (3)d(ij)k=∑i∑jd(ij)NijNk where

d(ij)k: distances between sub-samples (ij) and Cluster k

sis: distance of sub-sample i and cluster k

djk: distance of sub-sample j and cluster k

In this study, the number of clusters uses Euclidean distance. The distance between two points is calculated using the formula (4) (4)d(xi, xj)=∑z = 1p(xki-xkj)2 where:

d(x_i, x_j) : Euclidean distance between object i and object j

x_ki : object value to-i on variable k

x_kj : object value to-j on variable k

z : variable to z, z = 1, 2, 3, …, p

Dendrograms are mathematical and visual representations of the clustering procedure performed using hierarchical cluster analysis. The shape of a dendrogram is identical to that of a tree diagram. The points on the dendrogram represent Clusters, while the bar length represents the distance at which objects are combined in the Cluster. The dendrogram will be cut to find out the number of clusters formed from the longest difference. An example of a dendrogram shape is presented in Figure 1.

The analysis used to overcome the direct, indirect, or total influence of the predictor variables on the response variables in a model is called a path analysis. The method in path analysis is a development that requires the input of correlation coefficients between variables that have been previously analyzed [12]. To formulate hypothesis testing applied form of multiple regression analysis of the relationship between complex variables is one of them with path analysis. In path analysis, correlations between variables are related to the parameters of the model which are represented by path diagrams. The path analysis technique is a development of the correlation technique developed by Sewall Right. Path analysis is used if theoretically,

Cluster analysis aims to group objects so that between groups are heterogeneous and between objects in each group are homogeneous. Next, path analysis is performed by testing the path analysis assumptions. Furthermore, the estimation of the cluster integration model with path analysis is carried out. Cluster integration with path analysis is used to determine how much influence a group has directly, indirectly, or the total number of predictor variables (exogenous) on response variables (endogenous) in a model.

The Cluster integration model in path analysis is written in equations (5) and (6), if there are many Clusters 3: (5)ZY1=β11ZX1i+β12ZX2i+β13ZX3i+β14D1iZX1i+β15D2iZX1i           +β16D1iZX2i+β17D2iZX2i+β18D1iZX3i+β19D2iZX3i           +ε1i (6)ZY2=β21ZY1i+β22ZX1i+β23ZX2i+β24ZX3i+β25D1iZY1i           +β26D2iZY1i+β27D1iZX1i+β28D2iZX1i+β29D1iZX2i           +β210D2iZX2i+β211D1iZX3i+β212D2iZX3i+ε2i

The tests of normality can be sub-divided into two categories which are descriptive statistics and theory driven methods [14]. Skewness and kurtosis coefficients are categorized as descriptive statistics whereas theory-driven methods include the normality tests, such as Shapiro Wilk (SW), Kolmogorov-Smirnov (KS), and Anderson-Darling (AD) tests. The Kolmogorov-Smirnov test is arguably the most well-known test for normality.

Suppose that the sample consist of n ordered data points, x₁ < x₂ < … < x_n. The cumulative distribution of the data is a step function where the step is between k-1n and kn at each x_k. Drezner et al., [15] defined Kolmogorov test statistics as

with Φ(xk-μσ) is the cumulative normal distribution at x_k with given mean μ and variance σ²

Attitude is a person's personality that can influence the choice and action of something. According to Mar'at [16], attitudes are actions and behaviors that are determined by one element of personality toward an object with good or bad feelings. Individual perceptions of the attitude object will be influenced by two factors, namely internal factors or individual factors and external factors or external factors. Internal factors are physiological and psychological while external factors can be in the form of experience, knowledge, teaching, and learning process situations faced by individuals, norms that exist in society, obstacles, or drivers in society.

According to Ajzen [17], subjective norms are a person's beliefs about the approval of others for action. The other person is called a referent and can be a parent, friend, or someone who is considered important to someone. According to Fishbein and Azjen in Triastity and Saputro, subjective norms generally have the following two components:

a. Normative beliefs

Perceptions or beliefs about other people's expectations of themselves become a reference for displaying behavior or not. Beliefs are related to the opinion of a character or other person who is important and influential for the individual or the role model whether the subject should or should not perform a certain behavior.

Individual motivation to meet these expectations. Subjective norms can be seen as dynamics between the impulses perceived by individuals from the people around them and the motivation to follow their views (motivation to comply) in doing or not doing the behavior.

Goleman [18], Self-control is a careful and intelligent attitude in managing life, balance, and controlled policies, and the goal is to balance emotions, not suppress emotions, because every feeling has value and meaning. According to Calhoun and Acocella [19], there are three aspects of self-control, namely:

Behavior control (Behavior Control).

The compliance in question relates to the taxpayer correctly calculating the tax payable, paying taxes on time, and reporting all information on time. Jamin in Anisa reveals that the level of compliance of corporate taxpayers is better than individual taxpayers. This can happen because most of the corporate taxpayers use the services of a consultant in dealing with corporate tax problems, while the individual taxpayers take care of their own tax problems. The indicators used in this study which replicates from Bambang & Widi [20] are:

Tendency to behave obediently and pay

According to Budiono [21] behavior is a reaction or human action caused by an impulse that is seen from values, habits, driving forces, motives, and restraining forces as a person's response or reaction that appears. This is due to the experience of the process of stimulation and learning from the environment. According to Sulistiyono [22], obedience is an attitude that is willing to do whatever it is based on self-awareness or coercion that causes behavior according to what is expected or not. Obedient behavior is the interaction of individual, organizational, and group behavior. Obedience to pay is defined as a person's actions caused by self-awareness and coercion to pay their obligations.

Linkages used in this study are Ward, Average, and Complete linkages. The distance between clusters were measured using Euclidean distance. The best number of clusters were determined using a dendrogram. The following are the results of the cluster on each method.

From Table 1, it can be seen that there are 44 members in Cluster 1, 30 in Cluster 2, and 26 in Cluster 3.

Ward linkage method obtained as many as 2 dummy variables, where the dummy variable 1 will be worth 1 when the data set is in Cluster 2 and 0 for the other Clusters. The dummy variable 2 will be 1 when the data set is in Cluster 3 and 0 for the other Clusters.

From Table 2, it can be seen that there are 13 members in Cluster 1, 80 in Cluster 2, and 7 in Cluster 3.

Complete linkage method obtained as many as 2 dummy variables, where the dummy variable 1 will be worth 1 when the data set is in Cluster 2 and 0 for other Clusters. The dummy variable 2 will be 1 when the data set is in Cluster 3 and 0 for other Clusters.

Based on the resulting dendrogram, the clusters formed by the Average linkage method are 2 clusters. The table of the number of members in each cluster is presented in Table 3.

From Table 3 it can be seen that there are 46 members in Cluster 1 and 54 in Cluster 2. From the Average linkage method, it can be seen that the dummy variable is 1, where the dummy variable will be worth 1 when the data set is in Cluster 2 and 0 for the Cluster.

The integrated cluster model in the path analysis obtained is then separated according to each cluster.

The selection of the best linkage and model validity is by choosing the model that has the largest R² total, as in the equation, which can be briefly seen in Table 7 as follows.

By Table 7 The R² adjusted value of the Cluster integration path analysis with the Ward linkage method produces an adjusted value of 0.3158, meaning that the variables of attitude, subjective norms, and behavioral control, are able to explain the diversity of the variable intention to comply with paying by 31.58%, while the other 68.42% are influenced by variables outside the model. The R² adjusted value of the Cluster integration path analysis with the Ward linkage method produces an adjusted value of 0.8294, meaning that the variables of attitude, subjective norms, behavior control, and intention to comply with paying are able to explain the diversity of variables of obedient paying behavior by 82.94%, while the other 17.06% are influenced by variables outside the model. The coefficient of R² total determination of the Cluster integration path analysis model with the Ward linkage method is 0.8832.

The R² adjusted value of the Cluster integration path analysis with the Complete linkage method produces an adjusted value of 0.3268, meaning that the variables of attitude, subjective norms, and behavior control, are able to explain the diversity of the variable intention to comply with paying by 32.68%, while the other 67.32% are influenced by variables outside the model. The R² adjusted value of the Cluster integration path analysis with the Complete linkage method produces an R² adjusted value of 0.8230, which means that the variables of attitude, subjective norms, behavioral control, and intention to comply with paying are able to explain the diversity of variables of obedient paying behavior by 82.30%, while the other 17.70% are influenced by variables outside the model. The coefficient of R² total determination of the Cluster integration path analysis model with the Complete linkage method is 0.8808.

The R² adjusted value of the Cluster integration path analysis with the Average linkage method produces an adjusted value of 0.2841 meaning that the variables of attitude, subjective norms, and behavior control are able to explain the diversity of the variable intention to comply with paying by 28.41%, while the other 71.59% is influenced by the variable outside the model. The R² adjusted value of the Cluster integration path analysis with the Average linkage method produces an adjusted value of 0.8291, meaning that the variables of attitude, subjective norms, behavior control, and intention to comply with paying are able to explain the diversity of variables of obedient paying behavior of 82.41%, while the other 17.59% influenced by variables outside the model. The R² coefficient of total determination of the Cluster integration path analysis model with the Average linkage method is 0.8776.

Based on Table 7, the path analysis model with Cluster integration with the Ward linkage method has the greatest total determination value so that the path analysis with Cluster integration with the Ward linkage method is the best model. The total determination value of 88.32% is considered very good to describe the model.

From the results of the R² adjusted integrated cluster on path analysis with the Ward method, it has the highest R² adjusted value, where the Ward method consists of 3 clusters. For the Ward linkage method with 44 Cluster 1, 30 Cluster 2, and 26 Cluster 3. If the average variables for each Cluster are compared, it is found that most of Cluster 3 has the highest average value compared to other Clusters, so that Cluster 3 is a high Cluster. While Cluster 1 has the lowest average value compared to other Clusters, so Cluster 1 is a low Cluster. The mean value and standard deviation for each cluster are presented in Table 8.

Based on Table 8, it can be seen that most of the customers have a good attitude in the high and medium clusters, but in the low cluster, most of the customers have a fairly good attitude. For the subjective norm variable, most customers have good subjective norms in the high and medium clusters, but in the low cluster, most customers have fairly good subjective norms. For behavioral control variables, most customers have good behavioral control in high and medium clusters, but in low clusters, most customers have fairly good behavioral control. For the variable of intention to comply with paying, most of the customers have good intention to comply with paying in high and medium clusters, but in the low Cluster, most of the customers have a fairly good intention to pay compliance. For the obedient paying behavior variable, most of the customers have good paying obedient behavior in the high and medium clusters, but in the low cluster, most of the customers have good paying obedient behavior.

Testing the path coefficients simultaneously through the Testing of the Parameter Linear Function Hypothesis (HFLP) as described in section 8 with the following hypotheses:

Based on Tables 9, 10, it can be seen that the calculated F-value > F_{(0.05;2;98)(3.089)} resulted in the rejection of H0 so that it can be concluded that there is an influence on the path. There is a non-significant path on the intention to comply with paying and obedient behavior to pay with a dummy or not with a dummy.