Home / Current Issue / Paper 1719668
Bivariate Predictor Screening for Teacher Retention Intention in Urban Public Schools under Kenya’s Cross-County Deployment Policy: Evidence from Bungoma County
Subject area: Science,Engineering and Technology · Area of research: Applied Statistics
DOI: https://doi.org/10.64388/IREV10I1-1719668
Abstract
Teacher retention in urban public schools remains a persistent challenge in Kenya, driven in part by the cross-county deployment policy of the Teachers Service Commission (TSC). Before an individual-level predictive model of retention intention can be estimated, the field of candidate predictors must be reduced to those carrying genuine signal. This paper reports the bivariate screening stage of a larger modelling study. Fifteen candidate predictors drawn from three theoretical domains—teacher characteristics (Human Capital Theory), school factors (Institutional Theory), and policy variables (Push–Pull Theory)—were screened against a binary measure of teacher retention intention using data from 325 teachers in three urban sub-counties of Bungoma County (Bungoma Town, Webuye, and Kimilili). Each predictor was tested individually with Pearson chi-square tests for categorical variables, Welch independent-samples t-tests for ordinal and continuous variables, and simple binary logistic regression for unadjusted odds ratios; effect sizes were quantified with Cramér’s V and Cohen’s d. Applying the liberal retention threshold of p ≤ 0.25, eleven of the fifteen candidate predictors were retained. The four non-retained predictors were all categorical teacher characteristics (Gender, Marital Status, Subject Specialisation, and School Level), each with negligible effect size (V ≤ 0.069). The strongest associations were policy- and workload-driven: Teaching Workload (d = 1.190), Housing Adequacy (d = 1.041), and Transfer Application Status (V = 0.513; unadjusted odds ratio = 9.78). The three domains formed a clear ordering, with policy variables strongest, school factors intermediate, and teacher characteristics weakest. The retained set yields an events-per-variable ratio of 16.5, confirming that multivariate estimation can proceed without overfitting. The screening identifies housing inadequacy, geographic displacement, and heavy workload as the candidate risk factors most deserving of policy attention.
Keywords
Teacher Retention Intention, Bivariate Screening, Binary Logistic Regression, Cross-County Deployment
How to cite this paper
@article{1719668,
author = {Siakilo Rose Mukoya, Vincent Marani, John Sirengo},
title = {Bivariate Predictor Screening for Teacher Retention Intention in Urban Public Schools under Kenya’s Cross-County Deployment Policy: Evidence from Bungoma County},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {10},
number = {1},
pages = {985-994},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1719668.pdf},
abstract = {Teacher retention in urban public schools remains a persistent challenge in Kenya, driven in part by the cross-county deployment policy of the Teachers Service Commission (TSC). Before an individual-level predictive model of retention intention can be estimated, the field of candidate predictors must be reduced to those carrying genuine signal. This paper reports the bivariate screening stage of a larger modelling study. Fifteen candidate predictors drawn from three theoretical domains—teacher characteristics (Human Capital Theory), school factors (Institutional Theory), and policy variables (Push–Pull Theory)—were screened against a binary measure of teacher retention intention using data from 325 teachers in three urban sub-counties of Bungoma County (Bungoma Town, Webuye, and Kimilili). Each predictor was tested individually with Pearson chi-square tests for categorical variables, Welch independent-samples t-tests for ordinal and continuous variables, and simple binary logistic regression for unadjusted odds ratios; effect sizes were quantified with Cramér’s V and Cohen’s d. Applying the liberal retention threshold of p ≤ 0.25, eleven of the fifteen candidate predictors were retained. The four non-retained predictors were all categorical teacher characteristics (Gender, Marital Status, Subject Specialisation, and School Level), each with negligible effect size (V ≤ 0.069). The strongest associations were policy- and workload-driven: Teaching Workload (d = 1.190), Housing Adequacy (d = 1.041), and Transfer Application Status (V = 0.513; unadjusted odds ratio = 9.78). The three domains formed a clear ordering, with policy variables strongest, school factors intermediate, and teacher characteristics weakest. The retained set yields an events-per-variable ratio of 16.5, confirming that multivariate estimation can proceed without overfitting. The screening identifies housing inadequacy, geographic displacement, and heavy workload as the candidate risk factors most deserving of policy attention.},
keywords = {Teacher Retention Intention, Bivariate Screening, Binary Logistic Regression, Cross-County Deployment},
month = {July},
doi = {https://doi.org/10.64388/IREV10I1-1719668}
}