Factors influencing poor collection of revenue

Factors influencing poor collection of revenue

The respondents were requested to indicate the factors that amounted to poor collection of revenue within the county. Table 4.14 indicates that dishonesty of revenue collection clerks was a major factor resulting to poor revenue collection at 9.3%, 8.9% of respondents said it was poor sensitization on tax payment obligation, 8.1% dishonesty of taxpayers, 8.1% hostile taxpayers, 8. 1% poverty, 7.6% poor collection methods, 7.6% lack of motivation of revenue collection clerks, 7.6% strained economic actives due to non-performing economy, 7.6% lack of people with skills to deal with taxpayers, 7.2% lack of citizen participation in setting tax rates, 6.8% interference by external forces for their own interests, 5.5% poor accountability of taxes, 3.8% tax exemption, 3.4% limited tax base, and 0.4% setting high tax leading to tax evasion.

Table: 4.14 Factors Influencing Poor Collection of Revenue

4.6 Other determinants of revenue maximization in devolved units in Kenya

4.6.1 Respondents training on revenue collection

The respondents were asked whether they had taken training on revenue collection or not. The results in figure 4.8 showed that majority of the respondents (80%) had undertaken training on revenue collection while 20% had not. This shows that regardless of the majority having received training, there is variation as some devolved units do have untrained revenue clerks. The variation in the phenomenon is bound to reflect on the levels of revenue maximization. Ndunda et al. (2015) recommended that county governments needed to increase the competence of revenue clerks and other county officials by attracting skilled and competent employees for the purpose of increasing revenue collection.

Figure 4.8: Respondents training on Revenue collection

4.6.2 Department coordinating the sources of taxes

The respondents were asked to indicate which departments coordinate the sources of the tax collected. From the results presented in Table 4.15, County assembly was found to be the most common department that coordinates tax collection with 43.5% of the respondents, followed by line managers with 25.1%, followed by finance department with 23.6% and finally community barrazas with 7.7%. The Constitution also gives the Commission on Revenue Allocation (CRA) the responsibility for assisting county governments to tap into additional sources of raising their own revenue (COK article 209 (4, 5)). It is however vital for the right department with the right capacity to be involved in the coordination of these sources. Considering this variation in the responses to this question, the reflection of this variation is also expected on the levels of revenue maximization across the devolved units.

Table 4.15: Department coordinating the sources of taxes

4.6.3 Special training on how to deal with taxpayers

The respondents were asked to indicate whether they had undergone any training on how to deal with taxpayers. According to the results in Figure 4.9, 56% of the respondents agreed that the county had special training on how to deal with taxpayers whereas 44% disagreed. This shows that almost 50% of clerks across the devolved units have not undergone training on how to deal with taxpayers. Units with clerks who receive this kind of special training on handling taxpayers are expected to have better results in revenue collection. Empirical literature reveals that maximization of revenue is dependent on how the taxpayers are handled. (Lymer & Oats, 2010) recommends sending of reminder notices to taxpayers in a period of 2-3 weeks before taxes are due for collection in order to ensure that there are no tax arrears.

Figure 4.9: Tax Special Training

4.7 The influence of institutional capacity on revenue maximization

4.7.4 Descriptive statistics of institutional capacity on revenue maximization

The study sought to assess the situation of institutional capacity across the devolved units. The construct was assessed via proxy measurements that were formulated in questions that the respondents were asked in relation to the capacity of their institutions. The respondents were asked to state their level of agreement regarding each question as an ordinal measure of each dimension of institutional capacity. The summary statistics for the descriptive analysis of this construct were presented in Table 4.16.

Table 4.16: Descriptive statistics of institutional capacity on revenue maximization

The respondents were asked whether revenue collection staff sign and adhere to the business code of conduct. The results in the table indicated that 46.3% of the respondents strongly agreed to this, 36.6% agreed and 6.7% were uncertain. On the other hand, 6.5% strongly disagreed while 4.0% disagreed that revenue collection staff sign and adhere to the business code of the conduct. The mean response was 4.12 which is greater than 4 implying that on average, the respondents are in agreement. On whether adequate training offered to county staff result to accountability, integrity and transparency among revenue collectors, 43.8% of the respondents agreed, 40.5% strongly agreed, 8.0% were uncertain and 1.7% strongly disagreed. The mean response to this question (4.15) was also above 4 which showed that on average, the respondents agreed that when adequate training is offered to county staff, accountability, integrity and transparency among revenue collectors are addressed.

The sampled revenue clerks also responded to the question of whether revenue organization structure in place is efficient and effective. A majority (47.0%) agreed, 25.1% strongly agreed while 18.7% were uncertain. However, there were other 5.5% of the respondents who strongly disagreed and 3.7% disagreeing. On average, the clerks however considered to be in agreement based on the mean response which was about greater than 3 and almost equal to 4. The table also shows that the respondents averagely agreed that the county government has adequate resourced capacity in terms of different revenue collection methods and streams. The mean response to this question was 4.18 which is larger than 4 which is also supported by the majority(46.0%) who strongly agreed and 37.1% of the respondents who also agreed to this. Only 8.7% of the respondents were uncertain, 5.2% in disagreement and 3.0% in strong disagreement with the adequacy of resourced capacity in terms of different revenue collection methods and streams.

The majority (42.0%) of respondents were found to be in agreement with the assertion that they have an adequate internal control system for leadership and political goodwill and programs to empower citizen were in place. Of the responses received, 42.0% of the clerks strongly agreed, another 42.0% agreed respectively while 8.7% were uncertain. For those who were in disagreement, 4.5% strongly disagreed and 2.7% disagreed. The results support the mean response of 4.14 which shows that on average, the clerks were in agreement with the assertion of the existence of adequate internal control system for leadership and political goodwill and programs to empower the citizen. Finally, the respondents were also asked whether staff competency and motivation in their units was realized when the staff were empowered and given access to the latest revenue-enhancing and collection technology. A majority (44.8%) of the respondents agreed, 41.0% strongly agreed and 7.2% were uncertain. The remaining 3.2% 3.7% and disagreed and strongly disagreed respectively.

The responses on all the sub-dimensions (indicators) of institutional capacity, the average responses were above 3 (uncertainty). This implied that averagely the revenue clerks believe (agree or strongly agree) that the counties have in place or are striving to have in place adequate capacity for revenue collection. From empirical studies, these indicators have been suggested for the improvement of the institutional capacity.  Nyongesa (2014) recommended the development of a revenue management capacity by training personnel and establishing proper revenue management mechanisms, in order for the county to provide quality services to the people. Nyongesa also recommended decentralization of ICT based tax collection systems as a way of improving institutional capacity which was also considered in measuring institutional capacity in this study. Ndunda et al (2015) also gave similar recommendations to the indicators used in this study that county governments needed to increase the competence of revenue clerks and other county officials and attract skilled and competitive employees for the purpose of optimizing revenue collection.

4.7.5 Regression analysis on institutional capacity and revenue maximization

The goal of the study was to evaluate the determinants of revenue maximization in devolved units in Kenya. This was achieved by fitting a regression model for the primary data collected to determine the causal relationships between the determinants assessed in the study and revenue maximization. The regression models formed the basis for hypothesis testing that further informed the conclusions drawn on the specific objectives. Bivariate regression models were fitted to assess the direct influences of each determinant on revenue maximization followed by a multiple regression model for assessing the joint effect of the determinants on revenue maximization. In each regression model fitted, the R-square was used to determine the explanatory power of the model and the significance of the model tested using Analysis of Variance (ANOVA). The R-square is a measure of the variation in revenue maximization explained by the variations in predictors in the models fitted. T-tests were carried out on each regression coefficient estimates to determine the significance of each determinant. The adjusted R-square was also included in the model which is an adjusted form of the R-square to the number of predictors in multiple regression models. It is used in comparison to models with the different number of predictor variables included such that if additional variables are useful and significantly improve the model, the Adjusted R-square is bound to improve. However, if additional variables do not improve the model, the Adjusted R-square will not increase and would reduce if additional variables result in over-specification.

The first specific objective of the study was to examine the influence of Institutional capacity on revenue maximization in devolved units in Kenya. A simple linear regression analysis was fitted on this influence with revenue maximization as the dependent variable and the institutional capacity as the predictor variable. The regression analysis results were presented in Tables 4.17, 4.18 and 4.19.

In the model summary Table 4.17, The R and the R-squares are 0.146 and 0.021 respectively. The coefficient of determination shows that the variation in the predictor in this model which is institutional capacity explains up to 21% of the variation in Revenue Maximization. The remaining 79% of the variation in revenue maximization in this one-parameter model is explained by other factors that are not included.

Table 4.17: Model summary Table

An ANOVA was carried out for the regression model to assess the levels of variability within the regression model and inform general significance test on the whole model. The results as presented in Table 4.18, shows an F-statistic of 8.753 with a p-value of 0.003. The p-value is less than 0.05 implying that this bivariate regression model of institutional capacity on revenue maximization is generally significant. This gives a further implication of a significant causal relationship between the predictor in the model (institutional capacity) and the revenue maximization.

Table 4.18: ANOVA on institutional capacity

Further to the ANOVA test of model significance, a t-statistic was used to test the significance of the coefficient estimate of institutional capacity in the model (Table 4.19). The results revealed that the institutional capacity has a significant coefficient estimate (β= 0.040, t=2.959, p-value = 0.003). The P-value of the t-statistic to this estimate was less than 0.05 where this implies that the equation generated from this model is significant. The model generated the equation given below;

Table 4.19: Model Coefficient table for institutional capacity

The results for this bivariate model were used to assess and draw a conclusion on the objective of examining the influence of institutional capacity on revenue maximization in devolved units in Kenya.

H₀₁: Institutional capacity does not significantly influence revenue maximization in devolved units in Kenya

The p-value of the coefficient of institutional capacity in the model was found to be 0.003 which is less than 0.05. The null hypothesis was therefore rejected and a conclusion drawn that institutional capacity significantly influences revenue maximization in devolved units in Kenya. From this analysis, the significant coefficient estimate found to be 0.04 implies that increasing the levels of institutional capacity by one unit would result in an increase in revenue maximization in the devolved unit by 0.04. Building the capacity of institutions devolved within the devolved units in considerations of dimensions of public service values and principles, staff capacity, leadership and organizational structures would improve revenue collection. Findings from other studies also drew similar inferences. Otieno et al. (2013) found a correlation between effectiveness in revenue collection and Information Systems (IS) and efficiency. Another study by Nyongesa (2014) found and recommended among other factors improved staffing by personnel training as a key to the development of revenue management capabilities that would result in proper revenue management mechanisms.

4.8 The Influence of existing policies on revenue maximization

4.8.1 Descriptive statistics on the influence of existing policies on revenue maximization

Another independent construct in this study was the existing policies. The items (indicators) selected for this construct were also formulated into questions that were measured on an ordinal scale of 5. The respondents were asked to state their level of agreement with questions on existing policies in devolved units. The results were presented in Table 4.20.

Table 4.20: Descriptive statistics of the influence of existing policies

The results show that for the first question on existing policies, a majority (46.5%) of the respondents agreed that the citizens were aware of the role they play in the county, therefore, there was sufficient public participation in the county budgeting process, 37.6% strongly agreed while 6.5% were uncertain. Of the remaining respondents, 5.7% strongly disagreed whereas 3.7% disagreed. The average response was considered to be of an agreement by the revenue clerks that the citizens were aware of their roles and that there was sufficient public participation in the county budgeting process leading to maximization of revenue. The respondents were also asked whether the institutions for revenue collection were effective and vibrant when the county enacted policy on citizens to submit their taxes using database platforms. A majority (49.8%) were in agreement, 24.6% strongly agreed and 17.7% were uncertain. There were other respondents who did not believe this to be the case in their counties. These were 6.2% of the respondents who disagreed and 1.7% who strongly disagreed.

Another aspect measured was on the extent to which government policies and regulations affect revenue maximization. The majority (54.5%) of the respondents agreed that the government policies and regulations restrict the amount to be collected in the county and this depends on whether they are punitive or friendly, 27.1% strongly agreed, 9.2% were uncertain. Only 6.2% were in strong disagreement with the remaining 3.0% also disagreeing. As to whether revenue is maximized when line ministry set timelines for revenue collection in the local authority, 46.8% of the respondents agreed, 30.6% strongly agreed, 14.2% were uncertain, 6.2% disagreed and 2.2% strongly disagreed. On average, the revenue clerks agree with this assertion. This is shown by the mean response score of 3.9 which is greater than 3 (uncertainty) and tending towards 4. It was also observed that 43.3% of the respondents agreed that taxable income is maximized when policies in place ensure county staffs have received training and are competent in the collection of revenue. There were 33.1% of the respondents who strongly agreed, 14.7% who were uncertain while 5.0% and 4.0% disagreed and strongly disagreed respectively. These translated to a mean response score of 4 implying that on average the clerks believe that taxable income is maximized through this means. Finally, for this construct, the respondents were asked whether their counties have an existing business continuity policy in place. A majority (52.0%) of the respondents agreed to this, 31.3% strongly agreed, and 7.7% were uncertain. The remaining (4.7%) strongly disagreed and 4.2% disagreed. The mean response score for this question was 4.02 which indicated that the clerks’ average belief is that their counties have existing business continuity policies in place.

The results on the proxy measurements of existing policies revealed high scores on the levels formulation and implementation of policies in the devolved units. On average, respondents were in agreement with all the positive questions that sought to measure the levels of existing policies. The same items were also sought as measures of existing policies in empirical studies.  Nyongesa (2014) recommended citizen participation and formation of a tax database which were considered as measures of policies as essentials in revenue collection. Guldentops (2001) also cited the importance of citizen participation while Ndunda et al. (2015) recommended increased competence of revenue clerks and also investigated in this construct as key factors contributing to the optimal collection of revenue. From this descriptive analysis, it was observed that there are varying levels of existing policies from the various responses. Further analysis would reveal the effect of these variations on revenue maximization.

4.8.2 Regression analysis of the existing policies on revenue maximization

Another simple linear regression analysis was fitted aimed at the realization of the second specific objective which sought to determine the influence of policies governing revenue collection on revenue maximization in devolved units in Kenya. Tables 4.21, 4.22 and 4.23 show the results of the bivariate regression analysis between these 2 variables. Table 4.21 presents the model summary with the goodness of fit statistics of the model. The R which is equivalent to the correlation coefficient between the 2 variables is 0.211 while the R-square is 0.045. The R-square is a measure which shows that only 4.5% of the variation in revenue maximization is explained by variations of existing policies (the predictor) in the model. The remaining 95.5% of the variance on revenue maximization is explained by other factors that were not considered in this model.

Table 4.21: Model summary table on existing policies

Table 4.22 presents the ANOVA which consists of calculations providing information about levels of variability within the regression model. The information on variation due to regression and residuals were used to calculate the F-statistic that was used for testing the general significance of the model. The p-value was of the F-statistic found to be less than 0.05 which implied that the model is significant and the F-value (18.693) is greater than 0.05 indicating that the model is significant.

Table 4.22: ANOVA on existing policies

Table 4.23 shows the parameter coefficients of the model fitted. The results in the table include the coefficient estimates and the test statistics for investigating the significance of the estimated coefficients. The results show that “Existing policies which” was the only predictor in this model had a significant coefficient estimate (β= 0.044, t=4.324, p-value = 0.000) as shown by the p-value which is less than 0.05. The constant term in this model is also significant implying that the equation has an intercept and does not pass through the origin. The coefficient of existing policies is the expected change in revenue maximization due to a unit change in the levels of existing policies whereby 4.4% change in existing policy would bring about a unit change in revenue maximization. The intercept in this model is the expected level of revenue maximization when the levels of existing policies are perceived to be at zero. The equation generated from the model is given below;

Table 4.23: Model Coefficient table for existing policies

The analysis thus revealed that existing policies have a significant coefficient estimate in the model. The model results were used in testing the hypothesis linked to the objective seeking to determine the influence of existing policies governing revenue collection on revenue maximization in devolved units in Kenya.

H₀₂: Existing policies do not significantly influence revenue maximization in devolved units in Kenya

From the model, the coefficient estimate of existing policies was found to have a p-value of 0.000 which is less than 0.05. The null hypothesis was thus rejected and a conclusion drawn that Existing policies significantly influences revenue maximization in devolved units in Kenya. A unit positive change in the index of existing policies is expected to yield a 0.044 increase in the levels of revenue maximization. The inference drawn from the findings is that the formulation of policies and implementation of existing policies within the devolved units play a role in the maximization of revenue.

Further inferences drawn from this analysis are that the dimensions considered as indicators of policies in this study contribute to the levels of revenue maximization in the units. This is in line with empirical studies that also considered the development and implementation of policies as determinants of revenue collection. Marsden (1983) mentioned that change in tax policy will affect the economic planning and business activities of a country. A report on revenue collection in Kitui County (2016), recommended a citizen participation policy as one of the factors to consider to realize revenue maximization. The Public participation policy is also argued out by Common (2007) and (Chiumya, 2014). Public participation in public policy-making and policy implementation also keeps public officials in check (Chiumya, 2014). Common (2007), argued out, the essence of the commitment of public official and political support for successful implementation of public policy is hinged on public participation.

4.9 The Influence of legal frameworks on revenue maximization

4.9.1 Descriptive statistics of the existing legal frameworks on revenue maximization

Another specific objective of the study was to determine the influence of legal frameworks on revenue maximization in devolved units in Kenya. The respondents were asked to state their level of agreement with the regards to this particular objective. The results were presented in Table 4.24.

Table 4.24: Descriptive statistics of existing legal frameworks on revenue maximization

The results in the Table revealed that the respondents agreed that when a county has adequate mechanism to deal with corruption and cartels such as code of ethics, the county would be able to increase per capita income. This was indicated by a mean value of 4.23.  The majority (44.8%) of the respondents strongly agreed, 43.3% agreed and 5% were uncertain. The other 4% and 3% disagreed, strongly disagreed respectively. Regarding questions as to whether counties have adequate risk and compliance process in place, a majority (51.5%) of the respondents agreed, 34.6% strongly agreed, 7.7% were uncertain 4.7% disagreed and 1.5% strongly disagreed. The respondents on average (mean response of 4.13) agreed that when counties have adequate risk and compliance process in place then revenue collection increases.

The study also sought to examine whether there would be increased stakeholder participation in the event that all taxpayers and collectors understand existing laws and regulations. On this, Majority (53.3%) of the respondents strongly agreed while 35.6% agreed, 6.2% were uncertain, 1.5% disagreed and 3.7% strongly disagreed. The resultant mean response score of 4.33 was observed which indicated that on average, the respondents agreed that when all taxpayers and collectors in their counties understand existing laws and regulations there would be increased stakeholder participation.

The respondents were also asked whether revenue increases with the strengthening of existing institutions through automation like the e-sourcing system. A majority (46.5%) of the respondents strongly agreed to this while 37.1% agreed, 10.2% were uncertain, 3.7% disagreed, 2.5% strongly disagreed.  Finally, the respondents were expected to confirm or otherwise the assertion that when county employees adhere to the existing constitution, Acts of both Parliament and County Assembly in the discharge of their duties then revenue collection would be improved. The majority (45.5%) of them strongly agreed while 42.5% agreed and 5.5% were uncertain. There were 3.5% respondents who disagreed and another 3% who strongly disagreed to this.

It was observed that all the indicators of existing legal frameworks had their average responses above 3 (uncertainty). This implied that averagely, the revenue clerks believe (agree or strongly agree) that the counties are at least striving to observe and adhere to existing legal frameworks in hope of streamlining revenue collection. From empirical studies, these indicators have been suggested for the improvement of the institutional capacity. The Kitui county revenue collection report (2016) observed corruption as a challenge to revenue maximization among other factors. The Institute of social accountability, (2016) also notes the need for staff to be keen on enforcing compliance with corruption regulations. In the Constitution of Kenya (2010), legislative frameworks on revenue collection are formulated and should be adhered to by revenue collection employees. This study measured the levels of adherence to the existing legal frameworks in order to assess its effect on revenue maximization.

4.9.2 Regression analysis of existing legal frameworks on revenue maximization

The other specific objective sought in this study was to determine the influence of legal frameworks on revenue maximization in devolved units in Kenya. A bivariate regression model was thus fitted with revenue maximization as the dependent variable and the existing legal framework as the independent variable. Table 4.25 shows a model summary for this model which comprises of the goodness of fit measurement. The R and R-square are 0.691 and 0.478 respectively. This implies that 47.8% of the variance in revenue maximization is explained by variations in the existing legal framework and 52.2% by other factors not included in this model.

Table 4.25: Model summary table for the existing legal frameworks

The ANOVA carried out and presented in Table 4.26 show the information about levels of variability within the regression model and those of the residuals and test for the significance of the model. The results revealed that the model significantly predicted revenue maximization, F=366.062; p= <0.000. The p-value of the F-statistic is less than 0.05 which implies significance at 5% level and the F-value is also greater than 0.05 testing at 5% significance level using a one-tail test which implies that the model is significant. This further shows that the model and the variance explained by the regression predictor (existing legal frameworks) are significant.

Table 4.26: ANOVA on the existing legal frameworks

Table 4.27 showed the model coefficient Table. The results revealed that existing legal frameworks significantly influence revenue maximization. This was indicated by the significant coefficient estimate of existing legal frameworks (β= 0.884, t= 19.133, p-value = 0.000) at 5% level of significance as shown by the p-value that is less than 0.05. The constant term is also significant thus the equation generated from this model does not pass through the origin. The constant term is the expected level of revenue maximization at zero levels of existing legal frameworks as measured in this study. The model equation is given below.

Table 4.27: Model Coefficient table for the existing legal frameworks

The analysis revealed that when considered as the only predictor; changes in legal frameworks within devolved units influence the extent of revenue collection and maximization. The analysis results were used to test the third study hypothesis that was based on the objective that was to determine the influence of legal frameworks on revenue maximization in devolved units in Kenya.

H₀₃: Existing legal frameworks do not significantly influence revenue maximization in devolved units in Kenya.

The coefficient of existing legal frameworks in the model was found to have a p-value of 0.000 which is less than the 0.05 level of significance. The study thus rejected the null hypothesis and a conclusion was drawn that existing legal frameworks significantly influence revenue maximization in devolved units in Kenya. A unit improvement in the existing legal frameworks is bound to have an effect on the collection of revenue by 0.884. In order for county governments to maximize revenue, the existing legal structures and frameworks should be improved. A study by Gikandi and Bloor (2010) also inclined to similar results in relation to the adoption of e-commerce. They emphasized the role of Kenya Government in achieving a secure environment for e-banking activities by; putting in place clear laws, rules and regulations and providing relevant technical training to the regulatory authority to empower them to enforce the laws effectively. From this study, it is deduced that the implementation of existing regulations is key to revenue maximization in the devolved units. This supports the observations by Ronald (2011) that despite the fact that there are constitutional provisions for statutory allocations and internally generated revenues, local governments are tightly controlled and subordinated by state governments through sundry mechanisms, including manipulation of the disbursement of financial transfers to them.

4.10 The Influence of human factors on revenue maximization

4.10.1 Descriptive statistics of human factors on revenue maximization

The last specific objective of the study was based on the construct measuring human factors that affect revenue maximization in devolved units in Kenya. The study thus sought to first describe the situation regarding the human factors in the devolved units through other observable indicators measured on an ordinal scale. The Likert scale was used for the questions that sought to determine the level of agreement of the respondents to the questions on human factors. The summary statistics of the responses were presented in Table 4.28.

Table 4.28: Descriptive statistics of human factors on revenue maximization

For this construct, the study sought to find out whether the county revenue collectors possess the relevant skills. A majority (54.7%) of the respondents agreed, 24.4% strongly agreed, 13.9% were uncertain, 2.5% disagreed, 4.5% strongly disagreed. The mean response score of 3.92 indicated the respondents on average agreed that the council revenue collectors possessed relevant skills. Regarding the question as to whether the county staffs acquired training skill and education that matches their job, 51.2% of the respondents strongly agreed, 36.3% agreed, 6.5% were uncertain, 3.2% disagreed and 2.7% strongly disagreed. The average response was that the clerks were in agreement that the county staffs acquired training skill and education that matches their job as portrayed by the mean response score of 4.3. As to the statement that when employees working as revenue collectors are very disciplined then citizen participation is easily achieved. The majority (47.8%) of the respondents strongly agreed, 41.5% agreed, 5.0% were uncertain, 4.0% disagreed and 1.7% strongly disagreed. On average (mean=4.3), revenue clerks agree that when employees and revenue collectors are much disciplined then citizen participation and intensity will be realized. The study also sought to determine the levels of transparency and accountability of the revenue collection staff in the counties. A majority (46.0%) of the respondents agreed, 30.3% strongly agreed, 17.2% were uncertain that the revenue collection staff in the counties exhibit accountability and transparency. The other 4.2% and 2.2% disagreed and strongly disagreed respectively. On the last indicator of this study, most (50.7%) of the respondents agreed that their counties had a continuous staff training programs on revenue collection, 38.6% strongly agreed, 5.5% were neutral, 3.0% strongly disagreed and 2.2% disagreed. This resulted in a mean response of 4.2 which showed that revenue clerks in the counties agree that they have continuous staff training programs on revenue collection in their respective counties.

The items observed as human factors influencing revenue maximization have been observed and recommended by other studies. Kayaga (2010) cited the need for experienced personnel while noting that inexperienced and unqualified personnel are heightened by the lack of adequate training facilities and opportunities. Fjeldstad and Haggstad (2012) on the other hand recommended that measures are required to improve accountability of revenue collectors and elected officials. Apart from competence, Ndunda et al. (2015) also recommend the attraction of skilled and competitive employees for the purpose of increasing revenue collection performance. This study observed high levels of agreement by the revenue clerks to the questions asked indicating high levels of performances in these aspects of human factors measured. Variation was also noted thus giving foundation and ground for further analysis on the effects of human factors on revenue maximization.

4.10.2 Regression analysis of human factors on revenue maximization

A simple linear regression model was also fitted to assess the objective on establishing the influence of human factors on the maximization of revenue in devolved units in Kenya. The results were presented in Table 4.29, 4.30 and 4.31. Table 4.29 is a display of the model summary statistics which is for the goodness of fit. The R which shows the level of correlation between the 2 variables was found to be 0.376 while the variation coefficient (R-square) was 0.142. The results show that the variation in the model predictor (Human factors) explains up to 14.2% of the variation in Revenue Maximization. The remaining 85.8% of the variation in revenue maximization in this one-parameter model is explained by other factors that are not included.

Table 4.29: Model summary table for human factors

The ANOVA carried out for the regression model assessed the levels of variability within the regression model and inform of general significance test of the whole model. The results as presented in Table 4.30, shows an F-statistic of 65.965 with a p-value of 0.000. The p-value is less than 0.05 implying that this bivariate regression model of human factors on revenue maximization is generally significant. This gives a further implication of a significant causal relationship between the predictor in the model (human factors) and revenue maximization.

Table 4.30: ANOVA on human factors

Table 4.31 showed the model coefficient Table. The results revealed that human factors significantly influence revenue maximization. This was indicated by the significant coefficient estimate of human factors (β= 0.449, t=8.122, p-value = 0.000) at 5% level of significance as shown by the p-value that is less than 0.05. The constant term is also significant thus the equation generated from this model does not pass through the origin. The constant term is the expected level of revenue maximization at zero levels of human factors as measured in this study. The model equation is given below.

Table 4.31: Model coefficient Table for the human factors

The results in this bivariate regression model between human factors and revenue maximization informed the hypothesis test and conclusion on the last study objective that sought to establish the causal relationship between the 2 variables.

H₀₄: Human Factors do not significantly influence revenue maximization in devolved units in Kenya

The coefficient of Human factors in this regression model was found to be 0.449 with a p-value of 0.000. The p-value of this coefficient being less than 0.05 led to a rejection of the null hypothesis to conclude that human factors significantly influence revenue maximization in devolved units in Kenya. Improvement in the dimensions of human factors assessed in this objective such as accountability and transparency, employee discipline and acquisition of skills would significantly have a positive impact on the maximization of revenue. It was observed that when counties have continuous staff training programs on revenue collection then revenue collection is enhanced. The results, therefore, indicate that a unit increase in human factors would increase revenue maximization by 44.9%.

The findings are in line with theories and past empirical studies. According to Kayaga (2010), new technology alone is not enough unless the government recognizes the need for skilled tax officials. Effective tax administration requires qualified tax personnel with requisite skills and competencies to maintain these systems and operate them to their fullest potential. In this study, the results show that when employees working as revenue collectors are much disciplined then citizen participation and intensity will be realized. The revenue collection report in Kitui county (2016) on the other hand also recommended minimization of clerks (employees) handling cash to enforce accountability and citizen participation.

4.11 Correlation analysis

This section of the study sought to establish the significance, direction and strength of the linear relationship between revenue maximization, which is the dependent variable and institutional capacity, existing policies, legal frameworks and human factors which are the independent variables. This was achieved by performing a Pearson’s correlation analysis. Pearson’s correlation values range from −1 to 1. -1 indicates a perfect negative relationship, 0 indicates that there is no relationship between the variables while +1 indicates a perfect positive relationship. An absolute Pearson’s correlation values of 0.60 to 0.79 indicate strong correlation while those above 0.8 are very strong. Values 0.40 to 0.59 indicate moderate correlation, 0.20 to 0.39 weak correlation and those below 0.20 indicate very weak correlation or linear relationship between the variables (Asuero et al, 2006). The sign of the Pearson’s correlation coefficient value indicates the direction of the relationship.

The p-value of the correlation coefficient denotes the measure of significance of the relationship at a desired level of significance. This study tested for relationship between the variables on either direction at 5% level of significance ( ) and 95% confidence level thus considered a 2-tailed test basing the rejection criteria on a p-value being less than 0.025 ( ). Therefore, Pearson’s correlation analysis was performed in this study and the findings were presented in Table 4.32. All the relationships tested in this study yielded significant correlations with p-values less than 0.025. All the independent variables had significant moderate and strong relationships with revenue maximization. However, the study objectives sought to establish causal relationships that required further analysis to realize. It was however also noted that the independent variables also had significant correlations with each other. Very high correlations between independent variables could result in multicollinearity on models estimation thus a standard test for multicollinearity was therefore adopted before the variables could be used in regression models.

Table 4.32: Pearson’s Correlation analysis

4.12 Multiple linear regression

A multiple linear regression model was fitted to investigate the joint influence of the determinants studied; institutional capacity, existing policies, existing legal frameworks, human factors on revenue maximization (the dependent variable). Tables 4.33, 4.34 and 4.35 show the multiple regression analysis results.

Table 4.33 is summary Table for the multiple regression model which displays the goodness of fit of the model. The R-square in this model was found to be 0.654 which indicates that 65.4% of the variation in revenue maximization is explained by the variations of the predictors in this model. 34.6% of the variance of revenue maximization still remains unexplained by the joint effect model but by other factors that are not included in this study. The R-square to the number of predictors (4) in this multiple regression model shows that it is an improvement to all the individual bivariate models between each determinant and revenue maximization. The adjusted R-square in the multiple regression model is larger than all the adjusted R-square values from the bivariate models.

Table 4.33: Model summary table

An ANOVA was also carried out to test for the significance of the whole model. The ANOVA provides information about levels of variability within the regression model. The results as shown in Table 4.34 revealed that the model is significant F=187.632; p= <0.000. This implies that a significant variance in Revenue Maximization is due to the regression components since the F-value obtained is greater than 0.05 testing at 5% significance level using a one-tail test.

Table 4.34: ANOVA on revenue maximization

Further to the ANOVA which showed that the model is generally significant, implying a joint influence of the determinants on revenue maximization, t-tests were carried out on each coefficient estimate. The coefficient estimates Table 4.35 displays each coefficient estimate in the joint model with the T-test statistics. The resulting equation generated from the model is given below.

Table 4.35: Model coefficient summary table

The results of this multiple regression model in table 4.35 show that the determinants that were considered in this study jointly influence revenue allocation as shown by the significant value for ANOVA F-statistic that has a p-value of 0.0001 which is less than 0.05 testing at 5% significance level using a one-tail test. All the study variables were found to be significant since their p-values were less than 0.05 testing at 5% significance level using a one-tail test (Sig. <0.05). The study findings indicate that Legal Frameworks is the main factor that influences revenue maximization as given by 0.874 coefficients (p 0.000<0.05, t= 22.375) followed by human factors 0.344 (p 0.000<0.05, t= 9.060), then Institutional Capacity 0.072 (p 0.000<0.05, t=7.709) and finally existing policies as the least with 0.034 (p 0.000<0.05, t=4.974). The t-values also show the significance of the study since their value is greater than 1.96 for normal distribution at 5% significance level.

4.13 Secondary data analysis

The study also used secondary data to support results obtained from primary data. Secondary data was retrieved from existing records showing total revenue collected in each county and the sources of the revenue for the financial years 2014/2015, 2015/2016 and 2016/2017. The structure of the secondary data collected was multilevel longitudinal data with both cross-sectional across the counties and over time.

4.13.1 Descriptive statistics for sources of revenues for all the counties

The sources of revenue collected were taken as independent variables in the secondary data collected that were sought to affect the revenue collected. The sources studied were schools, hospitals, markets, parking and other unspecified sources. From the findings in Table 4.36, revenue from other sources account for 72% of the total revenue, hospitals accounted for 9.3%, parking had 14.1%, markets 4.2% and revenue from schools is only 0.3%. This shows counties should shift focus on other sources of revenue if they are to maximize revenue.

Table 4.36: Summary of the sources of revenue for all the counties

4.13.2 Mode of revenue collection

The study also considered the modes of revenue collection as predictors of revenue collected. Counties used different methods to collect revenue. In order to increase revenue, counties should use efficient methods of revenue collection. Table 4.37 below provides the average amount of money collected by all counties using different modes of collection. From the table, M-pesa was the most (55.25%) used the mode of collecting money with more than half the money collected via this method. Followed by 30% of the money collected that was by cash, then 8% by cheque and finally, RTGs had the least amount collected which was 7% of the total money collected.

Table 4.37: Mode of revenue collection

Figures 4.10 to 4.13 reflect the sources of revenue collected by counties. Sources of revenue per county examined and the results presented in a clustered bar for each county. It is observed that across the counties, other unspecified sources of revenue collectively form the larger proportion of total revenue collected. However, different counties maximize collection from different sources. Nakuru and Nyeri have had more collection from hospitals over the 3 periods with Nakuru maximizing this in the 2015/2016 period. Nairobi and Vihiga, on the other hand, collect more revenue from parking than other sources.

Figure 4.10: Sources of revenue in Nakuru County

Figure 4.11: Sources of revenue in Nairobi County

Figure 4.12: Sources of revenue in Vihiga County

Figure 4.13: Sources of revenue in Nyeri County

4.13.3 Total revenue collected

The revenue collection rate was considered the dependent variable measured across the counties over the period from 2014 to 2016. The actual total revenue collected and the budgeted revenue collected were used as indicators of the dependent variable. The study was interested in the variance which was generated as the difference between the budgeted and the actual revenue collected. The actual dependent variable used was then calculated as a ratio of the variance to the actual revenue collected. Table 4.38 shows a summary of the descriptive statistics on the ratio considering the multi-level structure. Then overall mean ratio was found to be -0.923 which shows that on average, the actual revenue collected is less than the budgeted and by 0.923 of the actual collection. This ratio has an overall standard deviation of 0.978 which shows the dispersion of the ratio across counties and over time. The variation between groups is the variation in the ratio across the counties and that variation within groups is the variation of the ratio over time within a county. The variation between groups is less than that over time implying more differences in the revenue collection rates within counties than is across.

Table 4. 38: Revenue collection rate

Figure 4.14 below shows the total revenue collected in Kisumu, Nakuru, Nairobi, Vihiga and Nyeri counties in the financial years 2014/2015, 2015/2016 and 2016/2017. The results show that Nairobi has always had high revenue collection than other counties over all the periods. It is virtually observable that each county has an almost constant amount of revenue collection over the period to imply minimal variation over time.

Figure 4.14: Total revenue collection per county

Figure 4.15 is a profile plot of the revenue collection rate calculated as a ratio of the actual budget variance to actual collection. The profile plot shows the revenue collection rates for the 16 counties studied over time. Each line represents a county. It is noted that the counties have very varying profiles of revenue collection rate but most have minimal changes in the rate over time. Some however show increases and decreases in the collection rates over time.

Figure 4.15: Revenue collection rate across counties over time

4.13.4 Panel data model specification

The secondary data collected was used to fit a regression model to assess how the revenue sources and modes of revenue collection influence the collection rate. Considering the structure of the dataset, tests for stationarity were carried out as displayed in Table 4.39. Each county had data observations for all the 3 time period yielding a balanced panel dataset.  Stationarity is visualized by fitting and assessing autocorrelation function curves (acf) and partial autocorrelation function curves (pacf) in time series datasets where stationarity is tested with statistical significance using the Augmented Dickey-Fuller (ADF) test. Panel data is collected as an array of cross-sectional groups of time series data thus the unit root tests in panel data considers and assesses stationarity across all these time series panels. The secondary panel data collected for this study was balanced thus the study used Hadri Lagrange multiplier (Hadri LM) stationarity test to assess the panel stationarity. This test investigates the null hypothesis that all panels exhibit stationarity which is rejected if the P-value of the Hadri LM statistic is less than 0.05. As shown in Table 4.39, the p-values of the statistic are all greater than 0.05 thus the study failed to reject the hypothesis of panel stationarity and concluded that the panel dataset exhibited panel stationarity.

Table 4. 39: Unit root / Panel stationarity tests

The model specification for the secondary data was based on tests to determine whether the data exhibits panel effects or not. In the absence of panel effects, the pooled model is adopted as the appropriate model specification. The absence of panel effects implies no heterogeneity caused by individual effects. A test was Lagrange multiplier (BP-LM) test was carried out to test whether the data exhibits random effects against a pooled model. In a random effect model, individual effects are assumed to exist but are uncorrelated to the model predictors. The p-value of the chi-square statistic is greater than 0.05 as shown in Table 4.40. The null hypothesis of the LM test is not rejected, indicating that a pooled OLS is better than a random effect model.

Table 4.40: Breusch and Pagan Lagrange multiplier test for random effects

Test:   Var(u) = 0

chibar2(01) =   1.96

Prob > chibar2 =   0.0808

The pooled model was then tested against a fixed effect model. Fixed effect model assumes the existence of heterogeneity (individual effects) and that the heterogeneity is correlated to the predictors. A Chow test was carried out to determine the significance of the fixed effects present in the data set. Based on the F-Statistic and the P-value provided above, the researcher fails to reject the null hypothesis that there are no significant fixed/individual effects (Table 4.41). This implies that a pooled OLS can be used instead of a fixed effects model which accounts for individual effects.

Table 4.41: Chow Test for Fixed Effects

From the Lagrange Multiplier test (random effects) and the Chow Test (fixed effects) provided above, the results indicate the there are no fixed effects or random effects present in panel data set. In this case, Hausman tests became irrelevant in selecting between the Fixed Effects model and a Random Effects model. A pooled OLS was be fitted instead as the appropriate model specification for the data (Table 4.42).

Table 4.42: Random-effects GLS Regression

4.13.5 Diagnostic tests

Like other regression models, panel data regression models are also based on classical assumptions. The classical assumptions of normality, autocorrelation, multicollinearity, homoscedasticity and time fixed effects were tested as given in Table 4.43.

The results from the time-fixed effects F-Tests are indicated below. The results omit the base year to avoid multicollinearity in the dummy variables for the years. From the F-statistic and the corresponding p-value reported, the null hypothesis of “no time fixed effects” is not rejected. Therefore no time-fixed effects are relevant for the analysis of this panel data set. The F-statistic and the corresponding p-value indicate that the null hypothesis of no first-order autocorrelation is rejected.

The results from Wooldridge’s test of autocorrelation by Wooldridge (2002) shows a violation of the non-serial correlation assumption. The p-value of the F-statistic was less than 0.05 implying the existence of a serial correlation of order one and a violation of the assumption. Homoscedasticity was tested based on Breusch-Pagan test which had a chi-square of 3.83 and a p-value of 0.050 which is greater than 0.05 implying no significant heteroscedasticity by the residuals that are thus concluded to be homoscedastic.

Normality was tested using the Jacque-Bera test for normality which is based on the skewness kurtosis approach considering a skewness of 0 and kurtosis of 3 to imply normality. The Jacque-Bera chi-square statistic had a p-value of 0.05 implying no significant deviation from normality. Multicollinearity was tested by computing the Variance inflation factors (VIFs) for the predictors which were all found to be less than 5. The average VIF as shown in the Table is 4.43 implying that the predictors did not exhibit multicollinearity.

Table 4. 43: Diagnostics of the pooled model assumptions

All the model assumptions tests were found not to be violated except for the assumption of non-autocorrelation. The model residuals were found to exhibit serial correlation thus the pooled OLS model could not be assumed to give the Best Linear Unbiased Estimates (BLUE). To adjust for the presence of this autocorrelation in the regression, we use robust standard errors in the Pooled OLS.

4.13.6 Pooled OLS Results

Given the diagnostic tests estimated and discussed above, the study sought to estimate a Pooled OLS which determines the effect of revenue source and mode of revenue collection of the ratio of the Variance to the Actual Revenue. Table 4.44 shows the robust pooled OLS model results that were fitted due to the violation of serial correlation. In this case, an independent variable that has a positive coefficient contributes adversely to the ratio of Variance to Actual Revenue (a higher ratio implies that the variance is increasing relative to the actual revenue).

Table 4.44: Pooled OLS: Linear regression

The results above indicate that the following sources of revenue contribute significantly (at 95% confidence interval) to the ratio of variance to actual revenue: Ratio of Schools Actual Revenue to Total Actual Revenue and the Ratio of Parking Actual Revenue to the Total Actual Revenue. The coefficients are both positive and can be interpreted as follows: An increase of 0.01 in the ratio of schools’ actual revenue to total revenue relates to a 0.76 increase in the ratio of variance to actual revenue. An increase of 0.01 in the ratio of parking actual revenue to total revenue relates to a 0.51 increase in the ratio of variance to actual revenue. Effectively, it can be noted that none of the sources of revenue has over the years contributed to a decrease in the variance from the budget. From the two significant variables discussed here, it can be noted that schools contribute more to the variance from the budget than parking revenues do.

Looking at the results from the mode of collection, it can be noted that only RTGS is positive and significant at the 95% confidence interval. The positive coefficient on this dummy variable implies that counties that use RTGS as a mode of revenue collection have, on average, increased ratios of the variance from the budget to actual revenue, indicating the detrimental nature of this mode of revenue collection. The other modes of collection, that is, cash, cheques and M-Pesa, albeit having insignificant coefficients, have a negative effect on the variance, indicating they reduce the ratio of variance to actual revenue (favourable).

Although the variables from the model are statistically significant, they result in an increase of variation if they are increased per unit. Cheque and M-Pesa have a negative effect which implies they result in a reduction in variation between actual and budgeted if their use is increased. Their use can only increase by considering the variables in the primary data by increasing institutional capacity, creating awareness using existing policies, adhering to the existing legal frameworks and taking human factors into consideration. Positive coefficients in the pooled regression model are also in tandem with factors influencing poor revenue allocation; lack of motivation of revenue collection clerks, poor collection methods and dishonesty of some of revenue collection clerks.

These results from the secondary data analysis support the objective findings from the primary data analysis. Adu-Gyamfi (2014) on revenue mobilization by districts assemblies in Ghana found that some of the complications undermining revenue mobilization are poor record-keeping on revenue sources. This is in line with the primary data analysis findings on objective 2 which found that changes in existing policies significantly influence and improve revenue maximization. Revenue sources were a dimension that was explored in the policies constructs which stated by Adu-Gyamfi (2014) requires proper record-keeping for adequate revenue mobilization and inclusion of such enforcements in revenue mobilization by laws and policies and exploration of new sources. The secondary data supports these findings as it shows from kept revenue records by sources that sources contribute heterogeneously to revenue collection rates. The sources such as schools and parking were found to significantly affect revenue collection while others such as schools hospitals and markets do not significantly affect revenue collection rates. Existing policies could be improved so as to exploit such impactful revenue sources in order to maximize revenue collection.

From the multiple regression analysis, the study found that the studied factors jointly influence revenue maximization. The Secondary data analysis supports this possibility as the dimensions of existing policies (revenue sources and collection modes) that are found to significantly influence revenue collection rate also support other independent constructs of the study (Institutional capacity and Human factors). This supports the indication that institutional capacity, human factors or both could be mediator variables to existing policies in the study model. This study, however, did not include the objective and proper methodology for examining mediation effects. This could be an area to be explored by future studies.

Table 4. 45: Summary of Hypothesis tests