**4. Experimental results**

**3. Methodology**

144 Leadership

effort, effectiveness and satisfaction.

data will be compared to other data mining algorithms.

This study will investigate the impact of three leadership styles (laissez faire, transactional and transformational) on the willingness of employees to exert extra effort, effectiveness and satisfaction. In this study, the researchers have opted for a Multifactor Leadership Questionnaire (MLQ) for collecting data. MLQ is a popular instrument for the study of leadership styles [24]. MLQ is a proven instrument for measuring the leadership styles since its inception in 1985. The study has employed a five-point Likert scale questionnaire to collect quantitative data (where 1 being strongly disagree and 5 being strongly agree). The data were collected from the employees of small- to medium-sized enterprises (SMEs) in the Bay of Plenty Region of New Zealand. A total of 24 organisations were selected to gather data samples (within the range of 200–225) for this research. The participants answered questionnaires that portray the style of leadership of their immediate level manager. The goal of this research was to find and compare leadership styles of managers and their implications from the employee's point of view. A total of 18 questions were designed to record the feedback on independent variables. Six questions each were designed to gather feedback on each leadership styles (transformational, transactional and laissez faire). The responses of these six questions were accumulated to find the score of each leadership style. Another nine questions were designed to gather responses of dependent variables. Three questions each for extra effort, effectiveness and satisfaction were designed, and the responses were calculated by taking the mean of each category questions. The pooled variable 'leadership outcome' was derived by taking the mean of extra

There are number of data mining algorithms available in the literature such as artificial neural networks, support vector machines, decision trees, rule-based algorithms, k-nearest neighbours and Bayesian algorithms [17]. However, in this research, three algorithms, namely OneR, J48, and Modlem are used. These algorithms are used in this research as the outcome of these algorithms are in the form of a decision tree or a set of rules, which are easy to interpret. OneR stands for One Rule or one level decision tree. This algorithm processes all input variables and selected single input variable that best model (classify) the output variable [30]. This is a very simple algorithm but has accuracy comparable to the state-of-the-art data mining algorithms. Mostly, OneR algorithm is used as a baseline algorithm against which the performance of other algorithms can be compared. Here, this algorithm will also be used as a baseline algorithm. The model accuracy and the degree of information extracted from the

J48 or C4.5 is a decision tree-based classification algorithm proposed by Quinlin [31]. J48 builds the decision tree on the training data using the concept of information entropy. At each node, the algorithm selects a variable that split the data into defined classes with highest information gain. The variable with highest information gain is select to expand the node. Modlem algorithm generates rules based on rough set theory and more information can be found in [32].

In order to run data mining algorithms, the dependent variables' score will be transformed into categorical variables. The numeric scores recorded on the extra effort, satisfaction, A total of 210 samples were collected, of which three were discarded as these had missing values and another five were removed after conducting outlier analysis. The final 202 samples were used to analyse the data. The reliability test was carried out to demonstrate the internal consistency of various measures used in this research. Coefficient alpha or commonly known as Cronbach's alpha was used to test the reliability of data, and the values of above 80% were obtained in all factors of this questionnaire.

#### **4.1. Descriptive analysis**

The total number of participants surveyed was 202 of which 66% were females and 34% were males. Approximately half of the participants (51%) were 25 to 34 years of age. The participants between the age 18–24 and 35–44 were 16% each. In terms of education, 36% of the participants had completed their bachelor's degree. There were 26 and 24% of participants with diploma and postgraduate qualifications. The remaining 14% had their higher secondary school completed. Participant's prior employment statistics showed that 44% of them had more than 7 years of total experience followed by 29% of them between 1 and 3 years of work experience. Participants with less than 1 year of experience and experience between 5 and 7 years were 7 and 8%, respectively. This depicts that the participants had fairly good experience of working in organisations. Working at the current job, more than half (52%) of the participants fall under the category of below 1 year of working at the current organisation, followed by 20% for 1–3 years of working at current job. Industries covered in this study ranges from health, hospitality, customer service, education and trades with health segment employing more females stood top at 34%, followed by hospitality at 25% and customer service 21%.

#### **4.2. Correlation and regression analysis**

**Table 2** presents pairwise correlation coefficients and the statistical significance of each coefficient showing the relationships between leadership styles and employee outcomes. The


**Table 2.** Pairwise correlation.

correlation coefficients suggest that transformational leadership has the largest score (0.71) on employee outcome of extra efforts and then on satisfaction (0.62) and on effectiveness (0.60). Higher scores in transformational leadership can lead to respectively higher employees' outcomes (extra effort, effectiveness and satisfaction). It is important to note that each of the correlation coefficients of transformational leadership with employee outcomes is statistically significant at 1% level. On the other hand, laissez-faire style has no statistically significant impacts on any of the employee outcomes. Interestingly, laissez-faire style reduces employee effectiveness. Transactional style has statistically significant impacts on employee outcomes; however, the correlation coefficients are much smaller in size compared to that of transformational. However, simple correlation coefficient is not a robust approach to predict such relationship because it is a simple two-way relationship and cannot control for all independent variables simultaneously and cannot control for other mediating factors that can potentially influence this relationship.

outcome (Model 1 in **Table 3**). The largest R2 value (55.8) in Model 1 also suggests that the model can explain almost 60% variation of the leadership outcome. In regards to specific outcome, transformational significantly increases efforts, satisfaction and effectiveness (pre-

**(Model 1) (Model 2) (Model 3) (Model 4) Leadership outcome Extra efforts Effectiveness Satisfaction**

Investigation on the Impact of Leadership Styles Using Data Mining Techniques

http://dx.doi.org/10.5772/intechopen.78660

147

(0.01) (0.03) (0.04) (0.04)

(0.01) (0.04) (0.04) (0.04)

(0.01) (0.05) (0.05) (0.06)

(0.04) (0.16) (0.15) (0.19)

(0.07) (0.29) (0.28) (0.30)

(0.01) (0.06) (0.05) (0.06)

(0.33) (1.14) (1.28) (1.19)

Laissez-faire style −0.01 0.01 −0.08\*\* −0.04

Transformational style 0.14\*\*\* 0.48\*\*\* 0.35\*\*\* 0.46\*\*\*

Transactional style 0.00 0.02 0.06 −0.07

Age −0.08\* −0.19 −0.17 −0.34\*

Gender 0.04 −0.15 0.35 0.22

Total service 0.01 −0.01 0.02 0.04

Constant 0.96\*\*\* 0.51 4.28\*\*\* 3.91\*\*\*

Observations 202 202 202 202 R-squared 0.558 0.519 0.383 0.406

On the other hand, transactional style has no significant impact on leadership outcome or on any specific outcome, whereas laissez-faire style seems to reduce effectiveness. It implies that if leaders are keen to increase employee satisfaction and effectiveness and to motive them to put their extra efforts, transformational style is the most effective. It is interesting to note that age is inversely related to leadership outcome and specifically to satisfaction; it might suggest that older people are more anxious if business environment is dynamic and changing and

In summary, in Section 4.1, we have presented percentage analysis on the demographical variables, which helped us understand the distribution of surveyed data on the basis

sented in Models 2, 3 and 4).

Robust standard errors in parentheses.

**Table 3.** Multiple regression analysis.

\* p < 0.1. \*\*p < 0.05. \*\*\*p < 0.01.

they feel less satisfied what inversely affect their outcome.

The multiple regression analysis is an appropriate approach better fit for such analysis. Multiple regression analysis was performed on the data, where the leadership outcome was regressed on leadership styles (transformational, laissez faire and transactional leadership) along with other mediating variables (e.g., participants' age, gender and total years of service). We also examined the relationship between specific leadership outcome (efforts, effectiveness, satisfaction) with leadership styles. The results are presented in **Table 3**. The multiple regression analysis between the leadership outcome and the leadership styles demonstrates that transformational leadership is the most effective style to bring most of the leadership outcomes.

More specifically, between the alternative choices of leadership styles, the coefficient estimates of multiple regression results suggest that, ceteris paribus, transformational style has the largest (0.14) and statistically significant impact (significant at 1% level) on leadership


Robust standard errors in parentheses.

correlation coefficients suggest that transformational leadership has the largest score (0.71) on employee outcome of extra efforts and then on satisfaction (0.62) and on effectiveness (0.60). Higher scores in transformational leadership can lead to respectively higher employees' outcomes (extra effort, effectiveness and satisfaction). It is important to note that each of the correlation coefficients of transformational leadership with employee outcomes is statistically significant at 1% level. On the other hand, laissez-faire style has no statistically significant impacts on any of the employee outcomes. Interestingly, laissez-faire style reduces employee effectiveness. Transactional style has statistically significant impacts on employee outcomes; however, the correlation coefficients are much smaller in size compared to that of transformational. However, simple correlation coefficient is not a robust approach to predict such relationship because it is a simple two-way relationship and cannot control for all independent variables simultaneously and cannot control for other mediating factors that can potentially

(0.70) (0.00) (0.00) (0.00) 6. Satisfaction 0.01 0.62 0.23 0.66 0.74 1

(0.86) (0.00) (0.00) (0.00) (0.00)

**1 2 3 4 5 6**

The multiple regression analysis is an appropriate approach better fit for such analysis. Multiple regression analysis was performed on the data, where the leadership outcome was regressed on leadership styles (transformational, laissez faire and transactional leadership) along with other mediating variables (e.g., participants' age, gender and total years of service). We also examined the relationship between specific leadership outcome (efforts, effectiveness, satisfaction) with leadership styles. The results are presented in **Table 3**. The multiple regression analysis between the leadership outcome and the leadership styles demonstrates that transformational leadership is the most effective style to bring most of the leadership outcomes.

More specifically, between the alternative choices of leadership styles, the coefficient estimates of multiple regression results suggest that, ceteris paribus, transformational style has the largest (0.14) and statistically significant impact (significant at 1% level) on leadership

influence this relationship.

1. Laissez-faire 1

146 Leadership

2. Transformational 0.11 1

Significance levels in brackets under each coefficient.

**Table 2.** Pairwise correlation.

(0.11) 3. Transactional 0.30 0.46 1

(0.00) (0.00) 4. Extra efforts 0.11 0.71 0.37 1

(0.11) (0.00) (0.00) 5. Effectiveness −0.03 0.60 0.32 0.67 1

**Table 3.** Multiple regression analysis.

outcome (Model 1 in **Table 3**). The largest R2 value (55.8) in Model 1 also suggests that the model can explain almost 60% variation of the leadership outcome. In regards to specific outcome, transformational significantly increases efforts, satisfaction and effectiveness (presented in Models 2, 3 and 4).

On the other hand, transactional style has no significant impact on leadership outcome or on any specific outcome, whereas laissez-faire style seems to reduce effectiveness. It implies that if leaders are keen to increase employee satisfaction and effectiveness and to motive them to put their extra efforts, transformational style is the most effective. It is interesting to note that age is inversely related to leadership outcome and specifically to satisfaction; it might suggest that older people are more anxious if business environment is dynamic and changing and they feel less satisfied what inversely affect their outcome.

In summary, in Section 4.1, we have presented percentage analysis on the demographical variables, which helped us understand the distribution of surveyed data on the basis

<sup>\*</sup> p < 0.1.

<sup>\*\*</sup>p < 0.05. \*\*\*p < 0.01.

of demographic variables (i.e. age gender, education level, etc.). In Section 4.2, multiple regression analysis along with Pearson correlation was performed to establish the significant relationships between leadership styles and outcome. These results demonstrated that transformational leadership style has statistically significant and positive impacts on extra effort, effectiveness and satisfaction and overall leadership outcome. In other words, in order to increase employees' productivity and organisational performance, managers should only pursue transformational leadership style.

#### **4.3. Data mining**

One Rule (OneR) is the first data mining algorithm used to analyse the data. This algorithm finds single best input attribute that split the data on the basis of output attribute (class label). The input variables used were leadership style scores (Laissez faire, transactional and transformational) and output variables were extra effort, effectiveness, satisfaction and leadership outcome (one at a time). The final rules generated on all four output variables (dependent variables) by OneR algorithm are stated in **Figure 1**. According to these results, the algorithm has found transformational leadership style the most influential in deciding the outcome of all four dependent variables. Moreover, these results strongly support the findings of the regression analysis carried out earlier in this research.

The rules generated by OneR algorithm are easy to interpret and understand. For example, the three rules obtained from the leadership styles (laissez faire, transactional and transformational—input data) and extra effort (output data) can be described as below:

**1.** If the transformational leadership score is less than 17.5, then employees will tend to exert low amount of extra effort.

**Figure 1.** Rules extracted using OneR algorithm.

**Figure 2.** The model (decision tree) obtained using J48 algorithm with an accuracy of 69%.

Investigation on the Impact of Leadership Styles Using Data Mining Techniques

http://dx.doi.org/10.5772/intechopen.78660

149


The accuracy of these set of rules was 66%, which means that of 202 data instances, 133 instances were classified correctly. The individual accuracy of all three rules is also provided in **Figure 1**, which can be read as correctly classified instances/total number of instances.

The accuracy values in the case of extra effort (**Figure 1**) can be read as: 66% of the information in the data can be extracted by following transformational leadership feature and three basic rules. Alternatively, it can be said that this model loses 34% of the information from the data.

J48 is the second data mining algorithm used in this research to extract actionable information. This algorithm builds a model in the form of a decision tree. The model built by this algorithm achieved an overall accuracy of 69% (140 correctly classified instances out of total 202 instances). In this experiment, three leadership styles were used as input data and leadership outcome variable was selected as a dependent variable. The final decision tree generated by J48 can be seen in **Figure 2**.


**Figure 1.** Rules extracted using OneR algorithm.

of demographic variables (i.e. age gender, education level, etc.). In Section 4.2, multiple regression analysis along with Pearson correlation was performed to establish the significant relationships between leadership styles and outcome. These results demonstrated that transformational leadership style has statistically significant and positive impacts on extra effort, effectiveness and satisfaction and overall leadership outcome. In other words, in order to increase employees' productivity and organisational performance, managers should only

One Rule (OneR) is the first data mining algorithm used to analyse the data. This algorithm finds single best input attribute that split the data on the basis of output attribute (class label). The input variables used were leadership style scores (Laissez faire, transactional and transformational) and output variables were extra effort, effectiveness, satisfaction and leadership outcome (one at a time). The final rules generated on all four output variables (dependent variables) by OneR algorithm are stated in **Figure 1**. According to these results, the algorithm has found transformational leadership style the most influential in deciding the outcome of all four dependent variables. Moreover, these results strongly support the findings of the

The rules generated by OneR algorithm are easy to interpret and understand. For example, the three rules obtained from the leadership styles (laissez faire, transactional and transforma-

**1.** If the transformational leadership score is less than 17.5, then employees will tend to exert

**2.** If the transformational leadership score is greater or equal to 17.5 but below 25.5, then

**3.** If the transformational leadership score is greater than or equal to 25.5, then employees

The accuracy of these set of rules was 66%, which means that of 202 data instances, 133 instances were classified correctly. The individual accuracy of all three rules is also provided in **Figure 1**, which can be read as correctly classified instances/total number of instances.

The accuracy values in the case of extra effort (**Figure 1**) can be read as: 66% of the information in the data can be extracted by following transformational leadership feature and three basic rules. Alternatively, it can be said that this model loses 34% of the information from

J48 is the second data mining algorithm used in this research to extract actionable information. This algorithm builds a model in the form of a decision tree. The model built by this algorithm achieved an overall accuracy of 69% (140 correctly classified instances out of total 202 instances). In this experiment, three leadership styles were used as input data and leadership outcome variable was selected as a dependent variable. The final decision tree generated

tional—input data) and extra effort (output data) can be described as below:

employees will tend to exert medium amount of extra effort.

pursue transformational leadership style.

regression analysis carried out earlier in this research.

will tend to exert low amount of extra effort.

low amount of extra effort.

by J48 can be seen in **Figure 2**.

the data.

**4.3. Data mining**

148 Leadership

**Figure 2.** The model (decision tree) obtained using J48 algorithm with an accuracy of 69%.

The first split of the data is made at and above values of 21 on transformational leadership feature. This tree generates overall 10 decision trees using combination of all three features. Careful interpretation of the decision tree suggested that transformational and transactional leadership styles were playing main role in the construction of the tree. Total of 187 instances were classified using these two variables. The laissez faire style data were responsible for allocating class labels to 15 instances. The accuracy of this decision tree was 69%, which means there was 31% loss of information in this model (decision tree).

Overall, low outcome (class label) data are on the left hand side of the tree and the high outcome data are situated on the right hand side of the tree with medium mainly covering centre of the tree. This coverage is aligned with the results of OneR algorithm (stated in **Figure 1**). According to **Figure 2** (marker B: Outlier), the node with transformational value less than or equal to 16 and transactional values greater than 19 can be ignored or considered as an outlier. There are total two instances associated to this node out of which one is with high outcome and other with either low or medium outcome. The marker C in **Figure 2** has an interesting behaviour. This node suggests that a relatively low transformational value can achieve high outcome. According to this node, transformational values between 17 and 21 (both inclusive), transactional values greater than 20 and laissez faire less than equal to 19 can also produce high outcome. This information can be seen as an alternative solution of getting high outcome without increasing the transformational leadership skill value to above 25 (as suggested by OneR and this decision tree).

The final data mining algorithm used in this study was Modlem. The algorithm has generated total of 72 rules from the data (input variables as leadership styles and output variables as leadership outcome). However, few selected rules are shown in **Figure 3**. This algorithm has achieved an accuracy of 93% (184 correctly classified instances out of 202 total instances). For clarity, rules are separated based on their class labels (low, medium and high). These rules provide a great insight into the data. For example, rule 59 states that if a leader will have a transformational leadership style score less than 16.5 and laissez faire score below 14.5, then employees will have low outcome. However, rule 60 states that if leader's transformational style score is below 14.5 and laissez faire score less than 18.5, the employees will again have low outcome. The information deduction by reading rule 59 and 60 is that even though some leaders have relatively higher laissez faire score (14.5–18.5) at the cost of decreasing transformational score, their leadership outcome status did not change (stayed low). Another way these rules can be used is devising various pathway strategies for the managers (leaders) to move from one low outcome class to higher outcome class. For example, a leader belonging to rule 60 with low leadership outcome can be suggested to move from low to medium outcome class by either following rule 1 or rule 11. In order to move from low to medium outcome class, leader must improve his existing laissez fair score by 3 or 4 points. Alternatively, leader can improve his leadership outcome by keeping the laissez faire score same and only improving transformational leadership style score from 14.5 to 21.5 (minimum 7 points). Similar strategies can be adopted to move employees from medium to high outcome class by identifying a rule that someone belongs to and then finding a most suitable transition rule in the next outcome class label.

It can be seen from the rules and decision tree of OneR and J48 that in order to achieve high outcome transformational leadership value of 25 or above must obtained. However, the rules extracted from Modlem algorithm in **Figure 3** suggest that a relatively lower value of transformation skills along with some specific values of transactional and laissez faire can also produce High outcome in employees (rules 35–38). Similar rules can be extracted for low and

Investigation on the Impact of Leadership Styles Using Data Mining Techniques

http://dx.doi.org/10.5772/intechopen.78660

151

**Figure 3.** Modlem algorithm obtained an accuracy of 91% (only selected rules are shown here).

medium outcomes.

The first split of the data is made at and above values of 21 on transformational leadership feature. This tree generates overall 10 decision trees using combination of all three features. Careful interpretation of the decision tree suggested that transformational and transactional leadership styles were playing main role in the construction of the tree. Total of 187 instances were classified using these two variables. The laissez faire style data were responsible for allocating class labels to 15 instances. The accuracy of this decision tree was 69%, which means

Overall, low outcome (class label) data are on the left hand side of the tree and the high outcome data are situated on the right hand side of the tree with medium mainly covering centre of the tree. This coverage is aligned with the results of OneR algorithm (stated in **Figure 1**). According to **Figure 2** (marker B: Outlier), the node with transformational value less than or equal to 16 and transactional values greater than 19 can be ignored or considered as an outlier. There are total two instances associated to this node out of which one is with high outcome and other with either low or medium outcome. The marker C in **Figure 2** has an interesting behaviour. This node suggests that a relatively low transformational value can achieve high outcome. According to this node, transformational values between 17 and 21 (both inclusive), transactional values greater than 20 and laissez faire less than equal to 19 can also produce high outcome. This information can be seen as an alternative solution of getting high outcome without increasing the transformational leadership skill value to above 25 (as suggested by

The final data mining algorithm used in this study was Modlem. The algorithm has generated total of 72 rules from the data (input variables as leadership styles and output variables as leadership outcome). However, few selected rules are shown in **Figure 3**. This algorithm has achieved an accuracy of 93% (184 correctly classified instances out of 202 total instances). For clarity, rules are separated based on their class labels (low, medium and high). These rules provide a great insight into the data. For example, rule 59 states that if a leader will have a transformational leadership style score less than 16.5 and laissez faire score below 14.5, then employees will have low outcome. However, rule 60 states that if leader's transformational style score is below 14.5 and laissez faire score less than 18.5, the employees will again have low outcome. The information deduction by reading rule 59 and 60 is that even though some leaders have relatively higher laissez faire score (14.5–18.5) at the cost of decreasing transformational score, their leadership outcome status did not change (stayed low). Another way these rules can be used is devising various pathway strategies for the managers (leaders) to move from one low outcome class to higher outcome class. For example, a leader belonging to rule 60 with low leadership outcome can be suggested to move from low to medium outcome class by either following rule 1 or rule 11. In order to move from low to medium outcome class, leader must improve his existing laissez fair score by 3 or 4 points. Alternatively, leader can improve his leadership outcome by keeping the laissez faire score same and only improving transformational leadership style score from 14.5 to 21.5 (minimum 7 points). Similar strategies can be adopted to move employees from medium to high outcome class by identifying a rule that someone belongs to and then finding a most suitable transition rule in the next

there was 31% loss of information in this model (decision tree).

OneR and this decision tree).

150 Leadership

outcome class label.

**Figure 3.** Modlem algorithm obtained an accuracy of 91% (only selected rules are shown here).

It can be seen from the rules and decision tree of OneR and J48 that in order to achieve high outcome transformational leadership value of 25 or above must obtained. However, the rules extracted from Modlem algorithm in **Figure 3** suggest that a relatively lower value of transformation skills along with some specific values of transactional and laissez faire can also produce High outcome in employees (rules 35–38). Similar rules can be extracted for low and medium outcomes.

In these set of experiments, three data mining algorithms, namely OneR, J48, and Modlem, were used to analyse the data. The experiments demonstrated that these algorithms can be used to better understand the data and how segments of the input data are associated with the output variable. The information extracted from these algorithms (decision tree or rules) can be used to devise better informed management strategies to improve organisational performance. The results obtained from our data mining algorithms support the claim of existing literature that transformational leadership style is most effective in organisations due to the fact that it inspires employees to perform better, increase employees' engagement and create a healthy environment where future business leaders can be nurtured [2].

> algorithms applied on the data (correlation, regression, OneR, J48 and Modlem). The outputs were correlation, regression and decision trees obtained using data mining algorithms. The feedback consisted of the information extracted from the regression analysis and various rules extracted from the data mining algorithms. Management can use direct and actionable output from various processes to direct their various internal stakeholders. In the light of this

[1] Daft RL, Pirola-Merlo A. The Leadership Experience. Boston: Cengage Learning,

[2] Hoon-Song J et al. Role of transformational leadership in effective organizational knowledge creation practices: Mediating effects of employees' work engagement. Human

[3] Eid J et al. Situation awareness and transformational leadership in senior military lead-

[4] Alloubani AM, Alloubani M, AlaDeen, Abdelhafiz IM, Abughalyun Y, Edris EEM, Almukhtar MM. Impact of leadership styles on leadership outcome (effectiveness, satisfaction and extra effort) in the private healthcare sector in Jordan. European

, Uswa Zahra<sup>1</sup>

Investigation on the Impact of Leadership Styles Using Data Mining Techniques

http://dx.doi.org/10.5772/intechopen.78660

153

, Chandan Ohri<sup>1</sup>

and

feedback, internal stakeholders can improve their productivity and performance.

\*, Muhammad Akhtaruzamman<sup>1</sup>

\*Address all correspondence to: waseem.ahmad@toiohomai.ac.nz

Resource Development Quarterly. 2012;**23**(1):65-101

ers: An exploratory study. Military Psychology. 2004;**16**(3):203

1 Toi Ohomai Institute of Technology, New Zealand

**Figure 4.** Components of the business information systems.

2 AGI Education Limited, New Zealand

Australia Pty Limited; 2009

Scientific Journal. 2015;**10**

**Author details**

Waseem Ahmad<sup>1</sup>

**References**

Binu Ramakrishnan<sup>2</sup>
