School of Economics and Finance, University of the Witwatersrand
A measurement note from ongoing MCom research using satellite night-light data
Night-time radiance observed from orbit and light experienced by a resident on their street are different quantities. Before using VIIRS radiance as a measure of public lighting in household-survey research, I test it against residents' own reported satisfaction with street lighting across more than 500 Gauteng wards in three rounds of the GCRO Quality of Life survey. The association is positive in every round and strengthens over time. This supports the proxy and gives a basis for judging how far attenuation bias is likely to pull estimates built on coarser geography toward zero.
Research that uses night-time lights as a measure of public lighting typically draws on NASA Black Marble VIIRS composites at roughly 500 m resolution. The instrument records light directed upward or scattered from surfaces. A resident experiences something narrower: whether the lamp on their corner is working.
If those two quantities are only weakly related, every downstream estimate is measuring something other than public lighting, and no amount of care in the specification repairs that. The question belongs at the start of an analysis rather than in an appendix.
Exposure is ward mean VIIRS radiance, extracted by zonal statistics over Municipal Demarcation Board ward polygons. The reference measure is resident-reported satisfaction with street lighting, taken from the Gauteng City-Region Observatory Quality of Life survey, which carries public ward identifiers. I compute survey-weighted correlations separately within each round rather than pooling, so that a change in the relationship over time remains visible.
| Survey round | Fieldwork | Respondents | Weighted correlation |
|---|---|---|---|
| Quality of Life IV | 2015/16 | 30,002 | 0.23 |
| Quality of Life V | 2017/18 | 24,889 | 0.26 |
| Quality of Life | 2020/21 | 13,616 | 0.52 |
Correlations are survey-weighted and computed within round across more than 500 wards. All three are positive.
Figure 1. The association holds in every round and roughly doubles by 2020/21.
A positive and consistent association supports treating satellite radiance as a stand-in for ground-level public lighting. It also does something more useful than passing a check. Classical measurement error attenuates coefficients toward zero, so knowing how closely radiance tracks lived experience at ward level gives a basis for judging how far estimates built on coarser geography are likely to understate the true relationship.
That matters because household panels are often geocoded only to a large administrative unit. Where the available geography is coarser than the ward level used here, the attenuation implied by these correlations is a lower bound on the measurement problem, not an upper one.
The strengthening across rounds is consistent with the later surveys coinciding with wider variation in whether municipal lighting works at all, though the design here cannot establish that.
These are weighted associations across wards, not within them, and a correlation does not validate a causal design. Satisfaction is self-reported and may respond to neighbourhood conditions other than lighting, so some of the association is likely picking up a broader service-delivery signal. Ward boundaries are also coarse relative to the lamp on a street. The exercise answers one narrow question, whether the exposure variable measures the thing it claims to, and should not be read as answering more.