Which national indicators explain life expectancy at birth? This project models 2016 life expectancy across 269 countries and regions using 21 other World Bank World Development Indicators, such as access to electricity, national income, school enrolment, health spending and mortality. It then tests whether average life expectancy differs between continents.
University of Essex, MA317 Modelling Experimental Data, group coursework (supervisors Dr Joseph Bailey and Dr Stella Hadjiantoni), November 2019. Authors: Dilpa Rao, Divya Arora, Rohini Subramaniam, Srinidhi Karthikeyan, Viraj Kumar Dewangan. My part was the missing-data imputation (Section 3).
- Data summary and plots: descriptive statistics for each indicator.
- Missing values: indicators with more than 50% missing values were removed, and the remaining gaps were filled by mean imputation. Section 3 of the report compares this with variable deletion.
- Collinearity: investigated with the Farrar–Glauber test, then predictors with high variance inflation factors were removed.
- Multiple linear regression: life expectancy regressed on the remaining indicators, selecting the model by R².
- ANOVA: tests the hypothesis that mean life expectancy is equal across continents.
| File | Description |
|---|---|
Group Coursework Report.pdf |
Report (17 pages), with the R code for every section in Appendix 7 |
LifeExpectancy.csv |
Dataset: 269 rows × 24 columns (country, code, life expectancy 2016, 21 predictors) |
The R code for each step is in Appendix 7 of the report: 7.1 summaries and plots, 7.2 imputation, 7.3 collinearity (mctest::omcdiag / imcdiag), 7.4 regression, 7.5 ANOVA. Two things to know before running it:
- The scripts read
LifeExpectancyFinal.csvfrom an absolute Windows path. That file isLifeExpectancy.csvwith a Continent column added for the ANOVA, and the extended version is not in this repository. - The imputation (7.2) is plain base R: drop columns with more than 50% missing values, then fill the remaining gaps with column means.
library(DMwR)is loaded but not used, so it can be omitted (DMwRhas been archived on CRAN).