CHAPTER FOUR

DATA PRESENTATION, ANALYSIS AND DISCUSSION OF FINDING

4.0 Introduction

This chapter presents analysis and interpretation of the empirical results of the study. This

comprises of preliminary estimation and empirical rules. The result for the Granger causality

and the Autoregressive Distributed Lag (ARDL) estimation was also presented.

4.1 Data Presentation

Table 1: Study Data

YEAR POV COR URT RGDPCPI

1991 -0.65393 1.098612 1.380779 12.31374
1992 0.346423 1.568616 1.368894 12.33377
1993 0.340749 1.098612 1.381784 12.28765
1994 0.356275 1.098612 1.378766 12.24359
1995 0.355574 1.098612 1.372449 12.21729
1996 0.35977 1.740466 1.360464 12.23312
1997 0.379805 1.098612 1.360977 12.23684
1998 0.380489 1.098612 1.35918 12.23717
1999 0.345715 0.641854 1.361233 12.21756
2000 0.406132 1.098612 1.348592 12.24048
2001 0.419368 1.098612 1.344951 12.27146
2002 0.385262 1.098612 1.322022 12.38725
2003 0.641854 1.029619 1.338154 12.43122
2004 0.852712 1.098612 1.332102 12.49274
2005 1.026042 1.098612 1.336841 12.5282
2006 1.131402 1.098612 1.33579 12.56007
2007 0.975691 1.308333 1.33263 12.5968
2008 0.795704 1.098612 1.330253 12.63506
2009 0.531804 1.098612 1.325217 12.68509

1
 2010 0.242162 1.098612 1.323088 12.73466
2011 -0.26007 1.098612 1.327075 12.75874
2012 -1.13943 1.098612 1.321756 12.77267
2013 0.779325 1.280934 1.308333 12.81028
2014 0.322808 1.098612 1.372195 12.84519
2015 -0.65393 1.131402 1.440072 12.84596
2016 -0.63111 1.163151 1.506075 12.80458
2017 -0.62736 0.916291 1.553714 12.78734
2018 -0.61249 1.916923 1.60201 12.78142
2019 -0.61249 1.774952 1.649812 12.77878
2020 -0.59421 1.824549 1.791593 12.73627
2021 -0.38566 1.458615 1.781036 12.74803
Sources: National Bureau of Statistics (NBS) Annual publications, Transparency

International (TI), World Bank, Global Hunger Index.

4.2 Data Analysis

Table 2: Descriptive Statistics

POV COR URT RGDPCPI

Mean 0.167884 1.210669 1.407995 12.53397
Medan 0.355574 1.098612 1.360464 12.56007
Maximum 1.131402 1.916923 1.791593 12.84596
Minimum -1.13943 0.641854 1.308333 12.21729
Std. Dev. 0.606031 0.28026 0.131049 0.237421
Skewness -0.44043 1.128234 1.902733 -0.11511
Kurtosis 2.114619 3.949736 5.486971 1.35145
Observations 31 31 31 31
Source: Author’s Computation using E-views 10.0 software

The summary of the statistics used in this empirical study is presented in Table 2, as observed

from the table Poverty (POV) has the lowest mean value of 0.167884 while Real GDP Per

Capita (RGDPCPI) has the highest mean value of 12.53397 where the mean value of

2
Corruption (COR) and Unemployment (URT) are 1.210669 and, 1.407995. Furthermore, the

table shows the maximum and minimum of POV as 1.131402 and -113943, while the

maximum and minimum of COR are found to be 1.916923 and 0.641854 respectively. The

maximum and minimum of URT and RGDPCPI are 1.791593 and 1.308333, 12.56007 and

12.84586 respectively. The analysis was also fortified by the value of the skeweness and

kurtosis of all variables involved in the models. The skewness is a measure of the symmetry

of the histogram while the kurtosis is a measure of the tail shape of the histogram. The bench

mark for symmetrical distribution i.e. for the skewness is how close the variable is to zero

while in the case of kurtosis, when it is three is called measokuitic but value lower them that

is called platykurtic and above is referred to as leptokurtic. The descriptive statistics table

shows that POV and RGDPCPI are negatively skewed while COR and URT are found to be

positively skewed.

4.2 Unit Root Test

It is generally known that the econometric estimation of a model on time series data demand

that the series be stationary as a non-stationary series usually result in misleading inferences.

Thus prior to the model estimation, a unit root test was conducted. A unit root test is a

statistical technique used to examine whether the series of two arm re variables are stationary

or not. There are many test used in determining the stationary of a variable, but this study

employed Augmented Dickey-fuller (ADF) unit root test for its variable testing of stationary.

The result for the ADF test was obtained as described below

3
Table 3. Augmented Dickey-fuller Result

Variable ADF Test Critical Decision

Value
5% value 10% value
1% value
Log POV -5.72489 -3.68919 -2.97185 -2.62512 I(1)
Log COR -3.65576 -3.67017 -2.96397 -2.62101 I(0)
Log URT -8.33372 -3.68919 -2.97185 -2.62512 I(1)
Log RGDPCPI -7.1451 -3.68919 -2.97185 -2.62512 I(1)
Source: Author’s computation using E-views 10.0 software

Table 3 from the analysis presents the stationary result for the variables. It shows that Poverty

(POV) was not stationary at level but after taking its first different at (1%, 5% and 10%)

respectively in all Mackinon Critical Percentage; Corruption (COR) was found to be

stationary at level at (1%, 5% and 10%) respectively in all Mackinon Critical Percentage. The

result further shows that Unemployment rate (URT) and Real GDP per Capita (RGDPCPI)

were stationary at fist difference. Since the Augmented Dickey Fuller test (ADF) values are

greater than the critical values at all level of significance, this implies that the variables of

interest are integrated at first order difference I(1), at level I(0), at first order difference I(1),

and at first order difference I(1) respectively, it indicates that unit root does not exists.

Therefore, which the order of integration of the variables we are at liberty to run ARDL for

our analysis.

Table 4: Pillips-Perron Unit Root Test Result

Critical Value
Variable ADF Test 1% value 5% value 10% value Decision
Log POV -8.26633 -3.67932 -2.96777 -2.62299 I(1)
Log COR -3.72741 -3.67017 -2.96397 -2.62101 I(0)
Log URT -3.36068 -3.67932 -2.96777 -2.62299 I(1)
Log RGDPCPI -6.97001 -3.68919 -2.97185 -2.62512 I(1)
Source: Author’s computation using E-views 10.0 software.

4
Table 4 above shows a Phillip-Perron unit root test, the result revealed that POV is stationary

at first difference I(1) at Bertlet-Kernel critical percentage (10%, 5% and 1%) respectively.

The Phillip-Perron unit root test also revealed that COR is stationary at level I(0). However,

URT and RGDPCPI were found to be stationary at first difference I(1). Since the Phillip-

Perron unit root test values are greater than the critical values at all level of significance, this

implies that the variables of interest are integrated at first order difference I(1), and also at

level I(0), at first order difference I(1) and at first order difference (1) respectively, it

indicates that unit root does not exists.

4.3 ARDL Result Presentation, Analysis and Interpretation

4.3.1 ARDL Long-Run Co-integration Test

The result of co-integration test is shown in Table 5 below. In order to know if our variables

are co integrated an equally give us the choice to run ARDL, we estimated our equation to

test the joint significance of the coefficients of lagged variables based on Wald test of F-test

with the aim of observing the long-run relationship among the variables. The decision rule is

that if the estimated F-test value is higher than the upper bound critical value, null hypothesis

will be rejected which signifies the presence of long-run relationship.

Table 5: ARDL bound test estimation result.

Model For Estimation F-Statistics Significance Critical bound F-

Level Statistics
I(0) I(1)
POV = f (COR,URT, RGDPCPI) 5.774024 1% 3.65 4.66
5% 2.79 3.67
10% 2.37 3.20
Source: Author’s Computation using E-views 10.0 software

Table 5 above reported the result of ARDL Co-integration which indicated that the

estimated F-statistics (5.774024) is greater than upper critical bound.

5
From the decision rule, if the computed F-Statistics is greater than the lower bound, then the

null hypothesis is not rejected and it concludes that there is no long run relationship among

the variables. Conversely, if the computed F-Statistics is greater than the upper bound value,

then there is a long-run level relationship. On the other hand, if the computed F-statistics falls

between the lower and upper bound values, then the results are inconclusive. Table 4 above

shows the results of the bound Co-integrated test. It demonstrated that the null hypothesis as

against its alternative is easily rejected at the 5% level of significance. The computed F-

Statistics of 5.774024 is greater than all the lower critical bound values at 10%, 5%, and 1%

respectively.

4.3.2 ARDL Long-run Coefficient Estimation

Table 6: Result of Estimated Long-run Coefficient Using: ARDL (1, 1, 0, 4,) Model

Variable Coefficient Std. Error t-Statistic P-Value

COR 1.409854 0.041496 1.675414 0.0121
URT 1.2387 1.550163 -0.79908 0.4353
RGDPCPI -2.06233 0.69395 -2.97188 0.0086
C 25.81099 7.762009 3.325298 0.004
Source: Author’s Computation using E-views 10.0 software

Table 6 reveals that the estimated long run coefficients of the selected ARDL (1, 1, 0, 4)

model are significant at 5% level of significance possessing expected signs. The coefficient

of Corruption (COR) is positive and significant at 5% level of significance, thus supporting

the contention that Corruption carry perceptible influence on Poverty. The positive

coefficient of Corruption (COR) of 1.409854 indicates that in the long run a 1% increase in

Corruption will leads to 140.9854 percent increase in Poverty, all things being the same. The

estimated coefficient of Unemployment (URT) is 1.2387 which is positive and insignificant

indicating that in the long run a 1% increase in Unemployment rate will bring an increase of

123.87 percent in Poverty rate in Nigeria. Moreover, the coefficient of Real GDP per capita

6
(RGDPCPI) is -2.06233 which is also negative indicating that in the long run 1% increase in

RGDPCPI decreases Poverty (POV) by 206.233 percent.

4.3.3 ARDL Short-run Coefficient Estimation

Table 7: Error Correction Estimation for Estimated ARDL (1, 1, 0, 4,) Model

Variable Coefficient Std. Error t-Statistic P-Value

D(COR) 0.525666 0.254681 2.064017 0.0546
D(RGDPCPI) 4.293111 2.558619 1.677902 0.1117
D(RGDPCPI(-1)) -0.89693 2.901351 -0.30914 0.761
D(RGDPCPI(-2)) 0.767786 3.06896 0.250178 0.8054
D(RGDPCPI(-3)) 8.656566 2.776503 3.117794 0.0063
CointEq(-1)* -0.76876 0.159228 -4.82806 0.0002
R-squared 0.558596 Mean dependent var -0.02748
Adjusted R-squared 0.453499 S.D. dependent var 0.490761
Source: Author’s Computation using E-views 10.0 software

The short run dynamics coefficients from the estimated ARDL (1, 1, 0, 4,) model are being

shown in table 7 where by the lag is selected by Akaike information criteria. The table shows

that the estimated lagged error correction term ECM (-1) is -0.76876 which is highly

significant at 5% level of significance and negative (ranges between zero and one) as was

expected having probability value less than 5%, which is 0.0002. These results support the

short-run relationship/co-integration among the variables. The feedback coefficient is -

0.76876 suggests that approximately 76.88% disequilibrium from the previous year’s shocks

in converge back to the long run equilibrium and is corrected in the current year.

The result further shows a positive and significant relationship between corruption and

poverty in the short-run. The coefficient corruption is 0.525666, meaning that a 1% increase

corruption will leads to 52.5666% increase in poverty in the short-run. Furthermore, the error

correction regression result reveals that there is a positive but insignificant relationship Real

GDP per capita and poverty in the short-run, where by a 1% increase in RGDPCI will result

to 429.3111% increase in Poverty in the short-run.

7
Nevertheless, the value of R-squared and Adjusted R-squared co-efficient of multiple

determinations stood at 0.558596 and 0.453499 respectively. This implies that 55 percent of

variation in Poverty (POV) is explained by the variation in all independent variables, while

the remaining 45 percent is accounted for by the error term (e t). The high value of R-squared

shows that the model is a good fit. This implies that the dependent variable considered in the

model does account to a large extend for the changes in the dependent variables.

4.4 ARDL Diagnostic Test

Table 8: Diagnostic test result

S/N TEST STATISTICS LM TEST P –VALUE

1 Serial correlation Chi-square (2) = 0.0769
2 Heteroscedasticity Chi-square (4) = 0.1303
3 Linearity F (1, 16) = 0.6266

Source: Author’s computation using E-views 10.0 software

The table 8 above shows result for ARDL post estimation diagnostic test for serial auto

correlation and heteroscedasticity as well as that of linearity of the model. The p-value for the

Breusch-Godfrey Serial Correlation LM Test is greater than 0.05 at 5% level of significance

with shows that is no evidence of serial auto correlation in the model. More so, the p-value

for Beusch-Pagan-godfrey heteroscedasticity test is greater than 0.05 at 5% level of

significance, hence, the model is said to be homoscedastic. In addition, RAMSEY RESET

test for linearity p-value is greater than 0.05 at 5% level of significance, showing that the

linearity of the model is well specified.

We conducted a diagnostic test as shown in the table above in order to ascertain the

efficiency and consistency of our model. The result revealed that the model passed all the

time series problems of serial autocorrelation, heteroscedasticty and normality.

Figure 1.Histogram Normality test

8
Source: Author’s computation using E- views 10.0 Software

The figure above shows a jarque-bera test was used to test for the normality of the data. It

shows that the P-value of 0.0000 is less than 0.05 at 5% level of significance. Hence it is

inferred that the data is not normally distributed.

4.5 F-statistics

It should be noted that the application of F-statistics is to determine the general statistical

significance and reliability of the regression model used for this study. The outcome of the

study depicts that the model used is statistically significant and equally reliable with the

coefficient of determination of (R2) of 55 % percent.

Decision Rule

If F calculated (F*) is less than the F- tabulated, we accept null hypothesis and conclude

the slope coefficient are simultaneously zero.

If F calculated (F*) is greater than F-tabulated, we reject null hypothesis and conclude the

slope coefficient are simultaneously zero.

Given our empirical F-calculated (F*) as obtained in the result of estimated regression is

5.774024, while that of F-tabulated at 5% level of significance V 1 = (k-1 = 4-1 = 3) and V 2

(N-k = 30-5 = 25), is found to be (2.61) which is less than 5.774024 for F*. Therefore, since

9
F-calculated (F*) > F-tabulated (i.e 5.774024 > 2.61) alternative hypothesis (H 1) is accepted

and concludes that all the slope coefficient is simultaneously zero.

4.6 Pair-wise Granger Causality

The idea of this test is to determine if the indices Poverty (COR, URT, and RGDPCPI)

provide significant statistical information about Poverty. Formality occur when there is a lift

in an index of poverty that leads to a later increase in poverty and vice-versa in establishing

Granger causality at threshold value of 5%, the P-value must not exceed 0.05 (p-value<

0.05). Table 5 (below) shows the outcome of causality test conducted on a two lag length.

Table 9: The Estimates of Pair-Wise Granger Causality Test Results

Null Hypothesis: Observatio F-Statistic P-value

n
COR does not Granger Cause POV 29 0.17811 0.8379
POV does not Granger Cause COR 1.90634 0.1705
URT does not Granger Cause POV 29 2.13208 0.1405
POV does not Granger Cause URT 1.38903 0.2686
RGDPCPI does not Granger Cause POV 29 2.68052 0.089
POV does not Granger Cause RGDPCPI 0.39445 0.6783
Source: Author’s computation using E- views 10.0 Software

The test for causality as provide by the table above shows that there is no causal relationship

between COR and POV. i.e COR does not granger cause POV as indicated by the p-value

(0.8379) also POV does not granger cause COR as indicated by the p-value (0.1705). The

result also shows a that there is no causal link between URT and POV as indicated by the p-

value (0.1405) also POV does not granger cause URT as indicated by the p-value (0.2686).

Furthermore, RGDPCPI does not granger cause POV as shown by the p-value (0.089) also

POV does not granger cause RGDPCPI as shown by the p-value (0.6789).

10
In summary, the granger causality test result revealed that there is no causal relationship

between all the variables under study.

11