INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 114
Logistic Regression Analysis of Effect of Perceived External
Determinants on Memebership Churn in Professional
Organizations in Kenya: A Case Study of the Kenya Institute of
Management
Ngetich Festus, Apaka Rangita
Department of Statistics and Actuarial Science, Maseno University
DOI : https://doi.org/10.51583/IJLTEMAS.2025.140300015
Received: 10 March 2025; Accepted: 21 March 2025; Published: 02 April 2025
Abstract: Membership churn brings about significant challenges to professional organizations, thus threatening their financial
stability and long-term sustainability. This study investigates the perceived external determinants influence on membership churn
at the Kenya Institute of Management using logistic regression. Specifically, the study evaluated the effect of economic
conditions, availability of similar services and industry and professional changes on membership churn. A cross-sectional
research design was employed and with data collected from 384 KIM members through structured online surveys based on quota
stratified sampling. The logistic regression model identified industry changes as the most significant predictor of churn (OR =
1.51, p < 0.001) followed by economic conditions (OR = 1.28, p = 0.012). Availability of similar services was found not to be
statistically significant (p = 0.071). Model evaluation by Hosmer-Lemeshow test (p = 0.8403) confirmed a good model fit
supported by Nagelkerke’s R² showing that 72.2% of churn variance was explained by the perceived predictors. The findings
suggest that professional organizations should strategically adapt to industry and professional shifts and economic fluctuations to
reduce member churn. Implementation of flexible payment plans and differentiation of services help mitigate churn risks. This
study enhances the understanding of membership dynamics in professional bodies and offering strategic insights to enhance
member retention in similar organizations.
Key words: Membership churn, Similar Services, Industry changes, Economic conditions, Logistic regression
I. Introduction
Membership churn is an enormous concern for membership-based institutions globally. Membership churn or attrition is the rate at
which members of member-based organizations discontinue their membership for a specified period usually annually.
Membership-based organizations such as professional associations, subscription-based services, and trade unions rely on a stable
and engaged member base for their financial sustainability and strategic growth. However, persistent membership churn poses as
a great challenge, where individuals discontinue their affiliation thus leading to decline in revenues, reduced organizational
influence and weakened community engagement. KIM will propose and implement targeted strategies that would promote
member satisfaction, member retention and long-term viability and sustainability in general from the study (Ahn, Han, & Lee,
2006). Traditional research on churn has focused extensively on internal factors like member satisfaction, service quality and
member engagement levels. However, external determinants of economic conditions, competition from similar services, and
industry and professional changes play a significant role in a member’s decision to renew or fail to renew their status at the end of
the current period. Unlike internal factors, external influences are beyond organization's direct control thus making them harder to
predict and manage.
Logistic regression is a statistical tool in modeling the probability of a binary result or two mutually exclusive outcomes dependent
on one or more predictor factors. The influence of these perceived external predictors on the likelihood that members terminate their
subscriptions at the end of their current term is the goal of this study thus the need for evaluation based on the ranks of their
impact member on churn.
The Kenya Institute of Management is a distinguished professional body in Kenya was established in 1954 to provide
personalized training, consultancy, membership services, Diploma and Certificate courses, and professional certification courses to
the general Kenyan population. KIM is renowned for its passion to professional development and management training to Kenya
firms and their employees. Despite KIM’s proactive approach the institution encounters immense challenges. Economic
fluctuations, competition from professional bodies offering similar services and evolving member expectations are contributors to
member retention rates’ fluctuation at KIM. This institutional context enables KIM's pivotal role to enhance professional
excellence and leadership in Kenya.
Statement of problem
Membership-based institutions like the Kenya Institute of Management, face persistent challenges in managing high churn rates
by threatening their financial stability, member engagement and long-term sustainability (Ahn, Han, & Lee, 2006). Despite
offering training, consulting and certification services, professional organizations continue to experience churn, suggesting that
current retention measures are insufficient. Existing research highlights determinants such as satisfaction, engagement and
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 115
economic conditions as key predictors of churn (Park & Ahn, 2022; Routh, Roy, & Meyer, 2020), yet data-driven insights
specific to Kenyan organizations remain limited. Furthermore, most studies rely on linear regression which is considered
inappropriate for modeling a binary churn outcome (Hosmer, Lemeshow, & Sturdivant, 2013). This study applies logistic
regression to identify and quantify the perceived external determinants influencing churn at KIM thus providing actionable
insights for targeted retention strategies.
Purpose of the study
The purpose of the study was to analyze and rank determinants associated with member churn at the Kenya Institute of
Management using logistic regression.
Specific objectives
i. To explore perceived external factors and their influence on membership churn.
ii. To fit an appropriate logistic regression model and rank predictors’ effects on membership churn.
Hypothesis
H
0
: There is no significant relationship between perceived
external factors and membership churn at KIM,
.
H
1
: There is a significant relationship between
external factors and membership churn at KIM,
≠
≠
≠
≠
≠
.
Scope of the study
This study aimed to identify the significant determinants associated with member churn at the Kenya Institute of Management.
The study investigated the predictor variables of the perceived external determinants’ effect on member churn across KIM's
membership base with 8989 members spread across diverse Kenyan regions in 13 branches. The study's geographic
scope
was
confined
to
KIM's
membership
department located
at
KIM headquarters South C, along Popo Road in Nairobi county. The
study employed stratified quota sampling techniques based on membership categories from 384 members. The study’s data was
collected using online surveys and descriptive and inferential data analysis. The study was conducted etween June to November
2024.
Limitation
The study was based on self-reported data from surveys which may have caused biases due to respondents’ recall and desire to
reveal true information. Furthermore, the external factors of economic fluctuations and industry changes could have emerged
unpredictably thus may have influenced the study's significance over time, but the study assumed no variation. Acknowledgement
of these limitations was critical in interpretation of results and recommendation on future research on membership dynamics and
organizational management.
II. Literature Review
Theoretical review
Logistic regression is a statistical technique used to make predictions of probability of a binary outcome influenced by one or
more predictor variables. Logistic regression is used when the dependent variable is of binary nature like success or failure, yes or
no, and churn or no churn. The model transforms the linear regression model's output to fit in a range of 0 to 1 using a logistic
function to allow for the moderate estimation of probabilities rather than the continuous values.
Odds
In categorical data analysis the odds indicate the ratio of likelihood of binary outcomes occurring to its likelihood of not
occurring. Equation (1) represents a ratio of the likelihood of success to the likelihood of failure/ not occurring.
Equation 1
Odds ratio
The odds ratio indicates the proportion of the odds of event occurrence in a group to those of its occurrence in another group.
Odds ratio measure shows the relationship/association between the exposure and outcome. Odds ratio quantifies the effect of the
independent variables on the dependent variable. The ratio indicates how the odds of the target of dependent event change with
respect to a one-unit increase in the value of predictor or independent variable.
The odds ratio for groups A and B is given as
.
The odds ratio concept interprets the results of logistic regression models. The odds ratio in the logistic function is e
β
j
where e is
the value of natural logarithm and β
j
is the weight of the predictor or independent variable,
.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 116
An odds ratio exceeding one signifies the event has a higher likelihood in occurrence as the predictor value increases, an odds
ratio of between 0 and 1 shows a lesser likelihood in occurrence as the predictor value decreases while an odds ratio equal of one
shows no association between the predictor and the predicted variables.
Logit function and logistic regression
The logit function is the natural logarithm of odds of target variable being 1 (churn) versus 0 (not churn). The logit function
transforms probabilities, lying between 0 and 1, into a continuous range of values from −∞ ∞, that follows a linear model.
The logit function is expressed as log
With p representing the probability of event under enquiry occurring P
(
Y = 1|x
)
.
The logit function is models the association between the binary outcome (target) variable and one or more independent predictor
variables.
Equation 2
By multiplying both sides by the antilog natural logarithm(e) in the logit equation and making p the subject we can derive the
probability of an event occurring as shown by equation(3).
we get;
Equation 3
MLE parameter estimation
The maximum likelihood estimators approximates the parameters β
0
, β
1
, … , β
k
values. The likelihood function represent the
possibilities of observed values based on model parameters β =
(
β
0
, β
1
, β
2
, … , β
k
)
.
Likelihood function for n observed values
L(β) is given by Equation 4;
Equation 4
The likelihood function for estimating co-efficients of the logistic model is thus found by substitution as;
The log-likelihood for the likelihood function,
(
)
is thus given as;
If we let
then;
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 117
Equation 5
To evaluate the estimates of parameters the log-likelihood function in equation 5 is differentiated partially to produce k
derivatives all with respect to the respective parameters and equated to zero as equation 6. These equations are then solved
simultaneously to get the estimates of
.
The partial derivatives of log L
(
β
)
with respect to each coefficient βj, kj=0,1,…,k) are;
Equation 6
Model Assumptions
There exists a linear association between perceived external predictor variables and logit of the outcome variables.
To ensure adherence of this the Box-Tidwell test was used. The predictor variable
interacts with its logarithm to form a term
.
We then fit a new logistic regression model;
Equation 7
If any
in equation 7 was found to be statistically significant the relationship between
and the logit was non-linear violating
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 118
the underlying assumption for linear association between predictor variables and the logit. Linearity association is only
ascertained when the p-value is greater than the level of significance, (Box & Tidwell, 2012).
There should be no multicollinearity among perceived predictor variables.
The variance Inflation Factor, VIF was used to quantify multicollinearity through measuring the extent to which one predictor’s
variance was inflated due to correlation with other predictor variables. For predictor variable,
, the VIF is calculated as shown
in equation 8;
Equation 8
With
been the co-efficient of determination found through regressing
on all other predictors. Multicollinearity is considered
extreme if the
, multicollinearity with
values between 5 and 10 shows moderate correlation that affect
regression estimates and
of between 1 and 5 shows low to moderate correlation between predictors. Low
multicollinearity coefficients are considered ideal.
Mode errorsl are uncorrelated with each other.
The Runs Test for Randomness by Wald and Wolfowitz (1940) is used to check whether residuals were randomly distributed.
Residuals showing patterns suggest correlation thus violating the assumption of independence.
Empirical Literature
Economic conditions are perceived to have a great effect on member retention rates. According to Park and Ahn (2022) economic
downturns frequently result in an increase in churn rates as members give priority to essential expenses over subscribing to
professional memberships. The analysis based on multiple professional organizations found out that in recession times members
in cost sensitive groups like entry-level professionals who were more predisposed to cancel and /or fail to renew their
memberships. The study showed that income level is negatively correlated to membership churn.
Routh, Roy, and Meyer (2020) examined the difficulty of anticipation of customer attrition in the times of competing risks. The
study presented a methodology incorporating competing risk analysis in random survival forest framework allowing more
accurate prediction of churn probability with no probability distribution assumptions. The study used data from the hospitality
industry showing that their approach was more effective than standard probability models with 20% gain in precision. This study
put emphasis on the need to accounting for competition in churn prediction during client retention tactics development (Routh,
Roy, & Meyer, 2020).
Changes in standard practices, procedures and trends substantially impact on membership churn in areas like telecoms where
customer expectations change more frequently. The study underlines that failure in reaction to these developments could
potentially lead to higher churn rates as customers seek better and more favorable options. The research on the Malaysian
telecommunications market revealed the efficacy of application of Net Promoter Scores (NPS) and data mining approaches to
determine and quantify customer turnover. The study discovered that customers with low NPS levels were more likely to churn
but with timely interventions of customer service improvement reduce the churn risk. The study also found out that the
classification and regression trees method produced the best accurate churn predictions indicating that organizations should
remain responsive to emerging industry developments to sustain customer loyalty and reduce churn (Moser et al 2018).
Social and political considerations of changes in government policy and social movements potentially have a significant effect on
membership churn. Giudicati, Riccaboni and Romiti (2013) illuminated that shifts in social norms and underlying government
rules have a tendency of altering customer expectations and behaviors resulting in increased churn if firms do not adjust
effectively. The advocacy of societal movements towards ethical business practices put pressure on corporations to adopt new
standards and failure to lead to customer churning. Changes in the government policy of new data protection requirements create
compliance issues impacting consumer trust and retention. According to the report firms should be responsive in adapting to
external influences to retaining clients and help in building long-term connections between members and the membership
organization (Giudicati, Riccaboni, & Romiti , 2013).
The study’s empirical literature demonstrates that all perceived external determinants all had significant impact on membership or
customer churn rates in professional organizations. The external factors of economic conditions, competition from other
professional bodies providing similar services and industry changes significantly affect churn rates necessitating responsive
strategies to mitigate the inherent effect of the associated determinants.
III. Methodology
Research Design
Creswell (2014) defines a research design as a guided plan aiding the researcher in planning and implementation of the study in a
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 119
systematic way that meets principle of validity and reliability of results. Mugenda and Mugenda (2003) states that descriptive
research encompasses data collection methods for testing the study hypotheses or answering study. This research on logistic
regression analysis of determinants associated with membership churn at the Kenya Institute of Management used the cross-
sectional research design. According to Saunders, Lewis, and Thornhill (2016) a cross-sectional study methodology is best fit to
examine how predictor variables of external factors influence membership churn. According to Creswell (2014) cross-sectional
studies aids provide a summary of relationships of target and predictor variables at a single point. The methodology allows for
thorough examination of complex interactions and causal pathways thus allowing KIM provision with insights for strategic
planning and retention tactics.
Area of the study
The research was conducted within Kenya with data collection made at KIM with members located in 13 KIM branches located in
various regions of the country. The geographic scope was significant for representative findings for the diverse membership base
of KIM.
Study Population
A study population is an enumeration of all elements contained in the entire group of individuals from which the sample is drawn
and about whom the researcher wants to make conclusions (Mugenda & Mugenda, 2003). The target population should contain all
the subjects or units meeting the specified inclusion criterion in the study. All the active and fully subscribed members of the Kenya
Institute of Management individual members in the year 2024 was the study’s target population. As of the records provided by
KIM’s membership department during research period, KIM had 8,989 individual subscribed members.
Table 1: Target Population
Member Category
Frequency
Percentage
Student
1535
17%
Associate
2354
26%
Full Member
5100
57%
Total
8989
100%
Source: KIM Membership Department (2024)
Sample and Sampling Techniques
A study sample is a small part of the population that is selected for collecting and analysis in a study under consideration (Mugenda &
Mugenda, 2003). The purpose of sampling is selection of a representative portion of the population whose statistics reflect the
broader population's parameters thus aiding in facilitation of study’s inferences about the population based on the sample
selection (Creswell, 2014). To achieve a representative sample in the study stratified random sampling technique was used. The
method involves division of a heterogeneous population of 8989 members into small homogeneous subgroups based on the
membership grade referred to as strata. In this study each membership category, student, associate and full members formed an
independent stratum and then random selection of elements from each stratum. This technique ensured that the different
population segments were adequately represented in the population thus enhanced the validity and reliability of the results
(Creswell, 2014).
Krejcie and Morgan (1970) formula was used evaluate the magnitude/size of the sample based on the finite target population of
8,989, the critical value of 1.96 at 95% confidence interval based on a normally distributed population, the perceived proportion
of 0.5 aiding in maximizing variability of the population and a marginal error of 5% (Krejcie & Morgan, 1970). The sample size
formula for finite populations is given by equation 9:
Equation 9
N is the number of elements in the population
Z is the standardized critical value at 95% confidence level (1.96)
p is proportion of the population likely to churn and 0.5 for maximum variation E is the marginal error of 0.05
Sample size, ≈ 384
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 120
The sample size for each stratum was obtained using proportion allocation as in equation 10. Proportional allocation ensures each
membership category in the study population is well represented accordingly in the sample effectively reducing bias and thus
increasing in precision of estimates. (Cochran, 1977)
Equation 10
With:
as the proportionally allocated sample size for s
th
stratum,
N
s
as population size of s
th
stratum and
N as size of population
n is size of sample
Table 2: Sample Size
Member Category
Frequency
Sample Size
Percentage
Student
1535
66
17%
Associate
2354
100
26%
Full Member
5100
218
57%
Total
8989
384
100%
Source: Researcher (2024)
Instrumentation
Creswell (2014) defines instrumentation as tools and methods used in collection of data in a research exploration. Tools
instrumentation encompasses the development, testing and the use of instruments such as questionnaires designed to gather
information from target participants (Mugenda & Mugenda, 2003). The goal of instrumentation is to ensure that the data collected
in the study are reliable and relevant to the research objectives (Creswell, 2014). In this study a questionnaire was used in data
collection. A questionnaire is a research tool that consists of questions designed to obtain information from respondents in the
study enquiry. The questionnaire consisted of closed-ended, open-ended and demographics aided in representative data collection.
Validity is the extent to how well a questionnaire as a research instrument captures what it is supposed to measure (Mugenda &
Mugenda, 2003). Validity is precision and trustworthiness of the data collection tool in capturing the genuine core of the construct
under study. Validity ensures that a study's findings are accurately reflecting the phenomenon investigated (Creswell, 2014). The
questionnaire was pilot tested with a small group of 40 KIM members. According to Mugenda & Mugenda (2003), a pilot study
should have 10% of the main study's sample size. Pilot testing aided in the detection of difficulties with question clarity, language
and structure as well as determining whether the questions effectively measure the constructs intended to measure.
Reliability is consistency or reproducibility of measurements thus a trustworthy instrument consistently measures what it is
designed to measure and produces similar results under consistent settings (Mugenda & Mugenda, 2003). High reliability results
indicate that shows that the measurement is consistent throughout time and in a variety of scenarios (Creswell, 2014). The
reliability of the Logistic regression analysis of determinants associated with Membership Churn at the Kenya Institute of
Management was measured Cronbach's alpha value. This statistical measure was used to examine the questionnaire's internal
consistency determining how closely a bunch of elements were related. A high Cronbach’s alpha value of more than 0.7 imply
good internal consistency meaning the items dependably assessed the same underlying concept (Tavakol & Dennick, 2011).
Data collection Procedure
The primary method of distribution of questionnaires in this study was online via google forms shared in net promoter score for
fully subscribed members during the month of September 2024 by the KIM membership department in collaboration with the
researcher. Study participants received a link to the online questionnaire completed at their own convenience. Members selected
who had limited access to the internet used paper-based questionnaires distributed during KIM events and meetings during the
study period. The completed questionnaires can be collected on-site or mailed back. Follow-up emails, WhatsApp messages and
short text messages reminders were used to promote members’ participation and ensured substantial response rate. Responses
were collected until all the 66 responses from student members, 100 from associate members and 218 from KIM full members
amounting to a total of 384 questionnaires were met in line with quota stratified sampling to ensure representativeness and
diversity, (Cochran, 1977).
Data analysis
Logistic regression evaluates the probability value of a binary outcome (Y) based on the associated predictor variables;
1
,
2
,
…,
. The outcome or target variable has two categories, 1 when the member churn and 0 when the member renews their
membership.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 121
Equation 2
With co-efficients β
0
, β
1
, … , β
k
being the model odds parameters for the Constant,
X
1
, X
2
, … , X
k
predictor variables respectively.
The logistic regression model for this study was as in equation 11.
Equation 11
Whereby:
1
is economic conditions,
2
is availability of similar services and
3
represents the changes in industry or profession.
Since the model is non-linear the logit link function was used to convert the function into a linear relationship in equation 12.
Equation 12
Statistical Significance of predictors
The p-value measure was used to determine the level of significance of the individual predictors in the logit model. P-value was
used to show the probability that observed association between the predictor and outcome variables occurred by chance.
The null hypothesis was rejected when p-value was less than the significance level of 0.05 at 95% confidence interval and showed
that the predictor significantly manipulated the dependent variable.
Wald test presented in equation 13 tested the significance of coefficients of predictor variables in the model.
Equation 13
The Wald statistics abide by the chi-square distribution with a single degree of freedom at 5% level of significance. The null
hypothesis was rejected when
The odds ratio acts as the transformation in odds of the outcome occurrence for a one-unit increase in the value of predictor
variable. A 95% confidence interval shows the range to which true odds ratio value is forecasted with 95% level of confidence.
An odd ratio for a given predictor Xi in logistic regression was computed as:
Equation 14
The null hypothesis,
while Alternative hypothesis,
. The CI for the coefficient is given by the limits;
Equation 15
Equation 16
The value of exponentiated bounds of the CI for the odds ratio was then given as;
at
confidence interval.
Evaluation of model fit
The Akaike Information Criterion (AIC) was used as an evaluation tool for comparing the goodness of fit across multiple
competing models in the study.
The AIC for each model was calculated as AIC = −2log(L) + 2k with L representing the model’s likelihood and k denoting the
model parameters count.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 122
−2log(L) measured the model’s fit with a lower value demonstrated a better fit while 2k is penalty for model complexity. The
model with the lower AIC was significantly the model of best fit.
Ethical Consideration
In scientific research study, ethical considerations preserve respondents’ rights and well-being and respect. All participants
provided informed consent before their voluntary participation in the study. This method provided clear and thorough information
on the study's purpose, procedures, possible risks and significance. Participants were given information that participation was
entirely optional, and they could have withdrawn their comments at any given
moment
without
penalization.
Participants
were
provided with
written
consent
as proof of agreement to participate in the study. The respondents' confidentiality and
identities were preserved during the study using unique random identifiers. The study followed general data privacy guidelines as
outlined in Kenyan legislation including the Data Privacy Act of 2019.
IV. Results and Discussions
Reliability Test
The reliability of scale of the data collection tool was conducted using the Cronbach alpha as shown in table 4.
Table 3: Reliability Statistics
Cronbach's Alpha
N of Items
0.81
11
Source: Research data (2024)
With the alpha of 0.81 the scale’s reliability was considered reliable. This showed an indication that while the 11 items were
reasonably consistent in measuring the same construct.
Descriptive Statistics
From the data analysis in table 5 it showed that 80.5% of members did not churn while 19.5% of respondent members churned thus
demonstrated most respondents remained members though with great variability. Respondents reported varying degrees towards
economic effect, 21.6% stated very significant effect and 18.2% reported no effect with mean score of 3.09 and a standard
deviation of 1.411 indicated economic conditions played a moderate to significant role in membership churn. Similar services
affected membership behavior where 20.1% reported very significant impact and 20.3% reported slight impact with mean score of
2.99 and a standard deviation of 1.418 thus gave a suggestion that competition played an influential role. Industry and
professional changes influenced membership with 22.1% of respondents reporting significant influence while 17.4% reported no
impact with mean score is 3.11 and a standard deviation of 1.404 showed that industry and professional shifts moderately
influenced membership churn.
Table 4: Descriptive Statistics
Variable
Indicator
N
Marginal Percentage
Mean
Standard
deviation
Churn
Not churn
309
80.50%
0.2
0.397
Churn
75
19.50%
Associate
100
26.00%
Full member
218
56.80%
Economic
conditions
Not at all
70
18.20%
3.09
1.411
Slightly
73
19.00%
Moderately
77
20.10%
Significantly
81
21.10%
Very Significantly
83
21.60%
Similar services
Not at all
77
20.10%
2.99
1.418
Slightly
78
20.30%
Moderately
76
19.80%
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 123
Significantly
76
19.80%
Very significantly
77
20.10%
Industry changes
Not at all
67
17.40%
3.11
1.404
Slightly
75
19.50%
Moderately
74
19.30%
Significantly
85
22.10%
Very significantly
83
21.60%
Source: Research data (2024)
Table 5: Case Processing Summary Table
Unweighted Cases
N
Percent
Selected Cases
Included in Analysis
384
100.0
Missing Cases
0
.0
Total
384
100.0
Unselected Cases
0
.0
Total
384
100.0
Table 6 shows the case processing summary for analysis that consisted of a total of 384 cases represented 100%. Missing and
unselected absences showed data completeness ensured data integrity.
Model Assumptions
Linearity of logits
Table 7: Box-Tidwell Test for Linearity of logits
Term
Estimate
Std Error
Statistic
P-Value
Intercept
-5.703
2.165
-2.634
0.008
Economic conditions
0.489
0.969
0.504
0.614
Similar services
0.613
0.984
0.622
0.534
Industry changes
1.046
1.004
1.042
0.298
Log Economic Conditions
-0.118
0.464
-0.254
0.800
Log Similar Services
-0.209
0.472
-0.443
0.658
Log Industry Changes
-0.305
0.479
-0.636
0.525
Source: Research (2024)
Linearity of logits was based on the Box-Tidwell test as presented in table 7. The analysis showed that all the interaction between
the continuous predictors and their logarithmic transformations was not significant at 5% level of significance. This showed no
evidence of non-linearity in the association between the continuous predictors and the logit thus satisfied the assumption of
linearity of the logit for the perceived predictors.
Test of multicollinearity
The test of multicollinearity in the logistic model was validated using variance inflation factor as shown in table 8. The VIF
values for the predictor variables ranged from 1.2 to a maximum of 2.33 that were all less than 5, indicating multicollinearity was not
a model’s significant problem.
Table 8: VIF Values
Predictor
VIF
Economic conditions
1.616434
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 124
Similar services
2.014424
Industry changes
2.300119
Source: Research data (2024)
Test of errors in independence
The runs test was used to rest independence of errors as shown in table 9. The test statistics of Z=0.048 were very close to zero and
suggested no presence of significant deviations form randomness while the p-value of 1 showed that the null hypothesis of error
sequence been random was not rejected further supported assumption of error randomness.
Table 9: Runs Test for Residuals Independence data: binary residuals
Standard Normal = 0.048, p-value = 1
alternative hypothesis: two-Sided
Source: Researcher (2024)
Logistic Regression Model Diagnostics
The full model was fitted to identify the relationship between member churn and the associated determinants as shown in
table 10. The hypotheses tested were.
H
0
:
There
is
no
significant
relationship
between
perceived
external factors and membership churn at KIM.
=
=
=
=
.
H
1
: There is a significant relationship between perceived
external factors and membership churn at KIM,
≠
≠
≠
≠
≠
.
Table 6: Binary logistic regression full model based on effect ranks
Predictor
Estimate
S.E
Wald
df
Odds Ratio
Odds ratio 95% CI
P-Value
Lower
Upper
Intercept
-4.050
0.591
46.92
1
0.02
0.000
Industry Changes
0.410
0.099
17.21
1
1.51
1.25
1.84
0.000
Economic Conditions
0.244
0098
6.26
1
1.28
1.06
1.53
0.012
Similar Services
0.182
0.010
3.25
1
1.20
0.99
1.49
0.071
Source: Research Data (2024)
The estimate for the constant was -4.050 with a Wald statistic of 46.92 and p-value of 0.000. The odds ratio was 0.071 with
the p-value less than 0.05 thus null hypothesis was rejected at the 5% significance level showing that the constant was significant.
The estimate for industry changes predictor was 0.410 with Wald statistic of 17.21 and p- value of 0.000. The odds ratio was found
to be 1.51 with 95% confidence interval of 1.25 to 1.84. Therefore, by virtue of the p-value been less than 0.05, the Wald statistic
greater than the chi-square statistic at 1 degree of freedom (3.841) and the predictor estimate not including the odds ratio of 1 the
null hypothesis was rejected giving an indication that industry changes had a significant positive effect on member churn.
The estimate for similar services was 0.182 with Wald statistic of 3.25 and p-value been
0.071. The odds ratio was 1.20 with 95% confidence interval of 0.99 to 1.49 thus making availability of similar services an
insignificant predictor supporting the p-value greater than 5%, Wald statistic of 3.25 that was less than chis-square statistic of 3.841
and the confidence limit having an odd ratio of 1 led to rejection of null hypothesis and thus similar services had no effect on
membership churn.
The estimate for economic conditions was 0.244 with corresponding Wald statistic of 9.231 and p-value of 0.000. The odds
ratio was 1.28 with the 95% confidence interval of 1.06 to 1.53. The p-value was less than 0.05, the Wald statistic (6.26) greater
than the chi-square statistic at 1 degree of freedom (3.841) and the predictor estimate not including the odds ratio of 1, then the
rejection of the null hypothesis at 5% level of significance indicated that economic conditions had a significant positive effect on
member churn.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 125
Evaluation of model fit
The model was evaluated using both the deviance statistic and the Hosmer-Lemeshow test. The decreasing deviance as each of the
predictor variables was added as indicated in table 8 showed that the model improved as an additional predictor was added into
the model. This was also supported by the small p-values at 95% confidence interval. Additionally, the Hosmer-Lemeshow test in
table 11 with
2
= 4.1822 and p-value of 0.8403 at 8 degrees of freedom showed that no existence of significant difference
between the predicted and observed values. This was validated by the level of significance (0.05) been less than the test p-value
(0.8403) showing no signs of misfitting.
Table 11: Hosmer-Lemeshow goodness of fit test
Hosmer and Lemeshow goodness of fit (GOF) test
data:
model$y, fitted(model)
X-squared = 4.1822, df = 8, p-value = 0.8403
Source: Research data (2024)
Table 127: Model summary
-2 Log likelihood
Cox & Snell R Square
Nagelkerke R Square
350.22
0.453
0.722
Source: Research data (2024)
The logistic regression model to predict membership churn had a -2 Log Likelihood of 350.22 demonstrating how well the
model fitted as compared to a baseline model that had no predictors as presented in table 12. The Cox & Snell R Square value of
0.453 suggested that an approximate 45.3% of the variance in membership churn was explained by the model's predictors and is a
measure does not reach a maximum value of 1. In contrast the Nagelkerke R Square adjusted for the maximum possible value
stood at 0.722, giving a reflection of about 72.2% of the variance in membership churn as explained by the model. This indicated a
strong fit and demonstrated that the model provided a substantial explanation of the variability in membership churn.
The Logistic Model Selection
Table 13: Deviance, Residual Deviance and AIC Values for Predictors
Predictors
Null
deviance
Deviance
Residual
deviance
Degrees of
freedom
AIC
Industry Changes
379.3
18.27
360.3
382
364.3
Economic Conditions
379.3
8.02
371.2
382
375.2
Similar Services
379.3
2.75
376.2
382
380.2
Industry changes, Economic conditions
379.3
-
353.5
381
359.5
Industry changes, Similar Services
379.3
-
356.7
381
362.7
Industry changes, Economic conditions, Similar
Services
379.3
-
350.2
380
358.2
Source: Research data (2024)
Table 13 presented the null deviance of 379.3 at 383 degrees of freedom represents the magnitude of unexplained variation in the
model when no predictors are included with only the intercept used as calculated with a base. This value serves as a benchmark
for comparing with other models that have some predictors included. A high null deviance values shows that the intercept-only
model does not explain variability to a high level in the outcome.
For each predictor added individually as in table 14 reported deviance shows the reduction in the overall deviance as compared
relatively to the null model. When Industry Changes is used as the only churn predictor, the deviance is reduced by 18.27 units
resulting in a residual deviance of 360.3 with 382 degrees of freedom. Thus, showing that the reduction quantifying the
improvement in fit when industry changes is included.
In contrast, when economic conditions are used alone the deviance reduces by 8.02 units resulting in residual deviance changing
to 371.2 while maintaining 382 degrees of freedom. Similarly, when Similar Services as the only predictor the deviance reduces
by 2.75 units yielding a residual deviance of 376.2 with 382 degrees of freedom. The residual deviance values indicate that on an
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 126
individual basis industry changes provide a stronger effect on improvement in model fit than economic conditions which also
perform better than similar services.
The Akaike Information Criterion is used to compare the models while penalizing models for model’s complexity. The model
with industry changes alone has an AIC of 364.3, economic conditions alone yield an AIC of 375.2 and similar services alone
gives an AIC of 380.2. When predictors are combined with industry changes the first to be included in the model since it has the
least AIC with economic conditions the residual deviance drops further to 353.5 with 381 degrees of freedom leading to the AIC
improvement to 359.5. Similarly, the combination of industry changes and similar services produced a residual deviance of 356.7
and a consequent AIC of 362.7. The full model all three predictors (industry changes, economic conditions and similar services
availability) achieves the lowest residual deviance of 350.2 with 380 degrees of freedom and led to the lowest AIC of 358.2. This
shows that the best model should comprise of all the perceived external determinants of churn.
Likelihood Ratio Test
From the results as in table 14 it was found that the reduced model without similar services predictor included had an AIC of
358.2 while that of the full model with the similar services predictor included had an AIC of 359.5. This yielded a chi-square
statistic of 3.302 with one degree of freedom and the corresponding p-value of 0.692. Since the p-value is greater than the level of
significance (5%), we fail to reject the null hypothesis and conclude that demonstrating that similar services as a predictor does
not significantly improve the model at 5% level of significance. Similar services are thus removed as it does not substantially
affecting the model’s explanatory power.
Table 14: Likelihood Ratio Test Results
Model
Residual DF
Residual
Deviance
AIC
DF
Change
Deviance/Chi-
Square Statistic
p-value
Decision
Reduced Model Without
Similar Services
381
353.5
359.5
-
-
-
Baseline
Model
Full Model with Similar
Services
380
350.2
358.2
1
3.302
0.069
Fail to Reject
H
0
From the logistic regression model outputs, the logit equation developed as presented equation 17.
Equation 17
V. Discussion of findings
The constant term represents the log-odds of membership churn when all predictor variables were set to zero. The odds ratio for
the constant was 0.02 representing the baseline odds of churning are very low when all predictor variables are at their reference
levels. This low odds ratio demonstrates that in absence of the effects of external predictor variables the likelihood of churning is at
its minimum.
For industry changes the odds ratio was 1.51 indicating that individuals affected by professional and industry changes are
approximately 1.51 times more likely to churn as compared to those not impacted. This shows that an increase in one unit in
economic difficulties leads to an increase in churn by 51%. This suggested that shifts and fluctuations within the industry and
professional lines have a significant effect on membership retention. With the high level of significance of p < 0.001 the changes
in industry and professional lines are a major external contributor of member churn, thus professional organizations should
closely monitor these changes to mitigate and reduce the inherent effects.
The odds ratio of 1.28 for economic conditions showed that worsening economic conditions increased the odds of churn by 1.28
times. This shows that an increase in one unit of odds ration leads that an increase in membership churn by 28%. Even though this effect is not as
strong as the effect of some of the other predictors is significant with p = 0.012 highlighting that broader economic challenges
significantly influence members' decisions to discontinue their memberships. Organizations should be wary of the underlying
macroeconomic environment and thus potentially offer flexible payment options during difficult times in order curb churn and
retain their members in the long run.
The odds ratio for the Similar Services determinant was found to be 1.20 with a high p-value of 0.820. This odds ratio
demonstrates that for every one-unit increase in the perception that similar services are available, the odds of a member churning
increase by 20% when all other factors are held constant. However, despite this positive relationship, the predictor is statistically
insignificant at the 5% significance level, showing that the observed effect could be due to random chance rather than a
meaningful relationship. This means that KIM should focus less primarily on external competition but rather on internal retention
strategies like member engagement, value differentiation and on external underlying economic conditions and changes in
professional and industry changes.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue III, March 2025
www.ijltemas.in Page 127
VI. Conclusions and Recommendations
Conclusions
Members affected by changes in their industries and/or professions are 1.58 times more likely to churn than those affected. This
signifies that external industry trends, external threats and professional shifts significantly influence members' decisions to renew
or fail to renew their annual memberships. KIM and other professional organizations should be agile towards adapting to industry
changes through provision of relevant services and resources thereby enabling members to easily navigate professional
transitions.
Worsening economic conditions increases the odds of churn by 1.28 times demonstrating that broader macroeconomic
determinants influence members' ability and willingness to renew their memberships. This suggests that economic fluctuations
and underlying personal financial constraints play a role in considerations of devising retention strategies and organizations should
allow their members to pay subscription fees on instalments.
Recommendations
Based on the study findings and conclusions on logistic regression of determinants associated with membership churn at the
Kenya Institute of Management the study recommends that in address the perceived external determinants of churn and improve
member retention membership institutions should differentiate those of its competitors by offering unique products and services and
put their emphasis on their distinct value proposition and they should endeavor to offer flexible payment plans during economic
downturns in order to support its member thus reducing the financial constraints driven churn.
Recommendation for further study
The study recommends a study on “Modelling membership retention over time in Kenyan professional organizations: A
longitudinal data survival analysis perspective."
References
1. Ahn, J. H., Han, S. P., & Lee, Y. S. (2006). Customer churn analysis: Churn determinants and mediation effects of
partial defection in the Korean mobile telecommunications service industry. Telecommunications Policy, 30(11), 552-
568. doi:https://doi.org/10.1016/j.telpol.2006.09.006
2. Box, G. E. P., & Tidwell, P. W. (1962). Transformation of the independent variables. Technometrics, 4(4), 531–550.
https://doi.org/10.1080/00401706.1962.10490038
3. Creswell, J. W. (2014). Research Design: Qualitative, Quantitative, and Mixed Methods Approaches (4th ed.). Thousand
Oaks, CA: Sage Publications.
4. Cochran, W.G. (1977). Sampling Techniques. 3rd Edition, John Wiley & Sons, New York.
5. Giudicati, G., Riccaboni, M., & Romiti, A. (2013). Experience, socialization and customer retention: Lessons from the
dance floor. Marketing Letters, 24, 409–422. doi:https://doi.org/10.1007/s11002-013-9233-6
6. Hosmer Jr, D., Lemeshow, S., & Sturdivant, R. X. (2013). Applied logistic regression (3rd Edition). John Wiley & Sons.
doi:DOI:10.1002/9781118548387
7. Kim, S., Gupta, S., & Lee, C. (2021). Managing Members, Donors, and Member-Donors for Effective Nonprofit
Fundraising. Sachin Gupta, Clarence, 85(3), 220-239. doi:https://doi.org/10.1177/0022242921994587
8. Krejcie, R. V., & Morgan, D. W. (1970). Determining sample size for research activities. Educational and Psychological
Measurement, 30(3), 607-610. doi:https://doi.org/10.1177/001316447003000308
9. Moser, S., Schumann, J. H., Wangenheim, F. v., Uhrich, F., & Frank, F. (2018). The Effect of a Service Provider’s
Competitive Market Position on Churn Among Flat-Rate Customers. Journal of Service Research,
21, 319-335. doi:https://doi.org/10.1177/1094670517752458
10. Mugenda, O. M., & Mugenda, A. G. (2003). Research Methods: Quantitative and Qualitative Approaches.
Nairobi: ACT.
11. Park, W., & Ahn, H. (2022). Not All Churn Customers Are the Same: Investigating the Effect of Customer Churn
Heterogeneity on Customer Value in the Financial Sector. Sustainability, 14(19).
doi:https://doi.org/10.3390/su141912328
12. Routh, P., Roy, A., & Meyer, J. (2020). Estimating customer churn under competing risks. Journal of the Operational
Research Society, 1138-1155. doi:https://doi.org/10.1080/01605682.2020.1776166
13. Saunders, m., Lewis, P., & Thornhill, A. (2016). Research Methods for Business Students (7th ed.). Harlow: Pearson.
14. Tavakol, M., & Dennick, R. (2011). Making sense of Cronbach's alpha. International Journal of Medical Education, 53-
55. doi:https://doi.org/10.5116/ijme.4dfb.8dfd
15. Wald, A., & Wolfowitz, J. (1940). On a test whether two samples are from the same population. The Annals of
Mathematical Statistics, 11(2), 147-162. https://doi.org/10.1214/aoms/1177731912
