# Employment and Household Poverty : KIHBS 2015/16 ## Finding Among people aged 15–64 classified as employed using the survey's native questionnaire rules, **29.7% lived in households below the supplied overall consumption-poverty line**. The approximate design-based 95% interval is **28.4%–30.9%**. This is a historical Kenya estimate for 2015/16, not a current rate, an individual earnings threshold or a causal effect of employment. The estimate is 34.7% for employed rural residents and 22.0% for employed urban residents. For contributing family workers, it is 40.6%; for own-account workers, 30.0%; for paid employees, 27.7%. These are separate descriptive domains, not adjusted comparisons. Household composition, geography and other factors can differ between groups. ## Employment reconstruction and validation Source: KNBS member questionnaire, printed pages 11–12 (PDF pages 12–13). A positive response to any D02_1–D02_6 work screen routes the respondent to the job section. Those not working in the last seven days can enter through D04 activity codes A–F and D07 return-to-work codes 1–3. D04 is alphabetic, not a numeric yes/no field. D07 codes 4–5 route to the job-search section. D06 is retained in the original data but the printed questionnaire does not make its yes/no answer an exclusion rule. This reconstruction reproduces **26 published values** in the Labour Force Basic Report, Table 3.3 (PDF page 35): the employment rate and employed total for each of ten age groups, all ages 15–64, rural residents and urban residents. All match at the report's one-decimal rounding, including 17,875.7 thousand employed and the 71.6% employment-to-population ratio. The earlier proxy based only on a populated primary-activity field differs for one working-age person. The questionnaire route is now the primary definition. Twenty-nine people aged 15–64 have missing work-screen responses and no positive employment route; they are unclassified for the poverty comparison. The published employment-rate denominator includes the full working-age population to reproduce the report. The non-employed poverty domain excludes these 29 respondents and must never be labelled 'unemployed'. This is the survey's **native historical definition**: it includes farming, unpaid family work, apprenticeship and volunteering. It is not a harmonised modern ILO employment measure. Excluding apprentices, volunteers and the other-activity category yields 29.7% at one decimal (29.68% unrounded, versus 29.67% with the native route). This sensitivity does not resolve the treatment of own-use agricultural production and is not a substitute harmonised definition. ## Poverty and weighting Household poverty is `padqexp < pl_abs`: supplied deflated monthly total consumption per adult equivalent compared with the supplied residence-specific overall poverty line. Retain the file's unrounded thresholds. Do not deflate `padqexp` again. Every member links to the same household welfare record through `(clid, hhid)` without row expansion or unmatched records. For a person-domain estimate, each member receives the household `weight` once. The numerator is the sum of these weights for employed people in poor households; the denominator is the sum for employed people. Do not use `weight_pop` on member rows, which would multiply household size twice. The consumption-based measure reflects pooled household living standards, not the worker's salary alone. The earlier national checks independently reproduced overall poverty (36.1%), food poverty (32.0%) and the working-age population (24,955.5 thousand). The current analysis uses 33,982 employed respondent records. Each domain's valid count and weighted totals are exported. ## Approximate sampling uncertainty The release contains 92 strata and 2,387 observed PSUs, at least 12 per stratum. The report describes a 92-stratum clustered design with 2,400 selected PSUs (PDF page 24); observed release coverage is not the same as the planned sample. Supplied final weights are retained, without inventing weights for absent clusters. Use the ultimate-cluster with-replacement Taylor approximation, using `strat` and `clid`. For domain indicator d and poverty flag y, p = sum(w*d*y)/sum(w*d). Sum w*d*(y-p)/sum(w*d) within each PSU; within each stratum, its contribution is m/(m-1) times the sum of squared deviations of those PSU totals from their stratum mean. Sum contributions over strata, then take the square root. All observed PSUs remain in the calculation, including zero contributions outside a domain. There are no singleton strata. Intervals use p ± 1.96*SE, bounded to 0–100%. These are approximate normal intervals, not official KNBS intervals. Finite-population corrections and later sampling stages are not specified in this implementation. Sampling intervals do not capture measurement error, classification uncertainty or nonresponse bias. No hypothesis test or causal comparison is claimed. Method references: [Stata survey estimation](https://www.stata.com/features/overview/survey-features/) and [subpopulation estimation](https://www.stata.com/support/faqs/statistics/subpopulation-estimation/). ## Reproduction and next stage Run `scripts/kihbs_work_poverty.py`. Evidence: `outputs/kihbs_work_poverty_validation.json`; aggregates: `data/processed/kihbs_work_poverty.csv` and SQLite table `kihbs_work_poverty`; executed companion: `notebooks/03_kihbs_work_poverty.ipynb`. Ten point estimates are independently recomputed in SQL from source expenditure, poverty-line and weight variables. Source files remain unchanged; no respondent records appear in exported results. Contract and benefit results use their question-specific eligible populations, as documented in the job-quality methods. Earnings remain excluded. The dashboard preserves the survey year and employment definition; KCHSP population estimates remain withheld pending usable weights and design guidance. Producer sources: [KNBS KIHBS catalogue](https://statistics.knbs.or.ke/nada/index.php/catalog/13), [member questionnaire](https://statistics.knbs.or.ke/nada/index.php/catalog/13/download/55), [Labour Force Basic Report](https://statistics.knbs.or.ke/nada/index.php/catalog/13/download/56), [Wellbeing Report](https://statistics.knbs.or.ke/nada/index.php/catalog/13/download/60). Archived local copies and their hashes are in the project source records.