New Public Lighting in Informal Settlements: A Field Experiment in Cape Town, South Africa

The intervention is a 10-watt, outdoor, wall-mounted, solar-powered LED floodlight, which comes with a 15-watt solar panel, installed on the outside of residents’ homes, typically above or near the front door of the house. The light is programmed to turn on automatically during evening civil twilight and to turn off automatically during morning civil twilight. Once the light turns on, it moves through a brightness schedule (or, dimming profile), which adjusts the brightness of the light throughout the nighttime hours to manage the battery and ensure the light can burn for an entire night on a single day’s charge. This light should provide better lighting than the existing 30-40-meter-tall high-mast lights for two main reasons. First, the light is installed at roughly the height of each structure (about 3 m), meaning light shines directly into public or semi-public spaces rather than reflecting off of roofs, which wastes light and creates strong shadows for pedestrians. Second, the lights are distributed throughout the settlement, providing illumination in areas that may be situated far from the high-mast lights and thus, receive little to no light at all.

Households receive the light free of charge. Randomization is done at the path segment level and compound level (see the Experimental Design section). Compounds can be thought of as cul-de-sacs or shared semi-private spaces within the settlement that are used by several households, but are often gated. Every household that has a front door facing a selected path segment or compound receives a light during the intervention phase. At the end of the study, all households in the informal settlement will receive a free light.

Note: The idea for the use of wall-mounted lighting for public lighting in an informal settlement was developed as part of the doctoral work of ETH Zurich Institute for Science, Technology, and Policy (ISTP) PhD student Stephanie Briers who is pursuing a PhD in architecture.

This field experiment is intended to test the efficacy and impact of wall-mounted, outdoor, solar LED lighting as an alternative form of public lighting for informal settlements. We focus on one informal settlement in Cape Town. The light will be installed above or near the front door of each house on a treated path segment or compound so that it beams light into the public space (pathway or shared compound). The light itself is a standard 10-Watt, outdoor, solar light, which we have modified to meet the public lighting needs of residents in informal settlement and to make the lights more resilient to vandalism or theft. Working with collaborators, we developed and included the following customizations to the standard solar LED lighting model:

- Increased the size of the battery to ensure that the light could store enough solar energy to be on for at least 14 hours every evening. We calculated that 14 hours is the maximum number of hours on the darkest day of the year that someone might need public lighting in order to navigate. The battery is now made up of two 4500mAH lithium ion 18650 batteries.

- Increased the size of the solar panel from 8 or 10 Watts, which is more common for a 10-Watt light, to 15 Watts. This adjustment is to ensure that the battery will fully charge, even in sub-optimal weather.

- Battery life in a solar light is also managed by a brightness schedule or dimming profile, a schedule which dictates how bright the light shines at different points throughout its nightly lifecycle. We adjusted the dimming profile of the solar light to more closely match the darkest hours in Cape Town that tend to have the most pedestrian traffic (e.g., peak commute times). In addition, this scheduling enables more efficient use of the battery to ensure that each light will stay on for an entire cycle on a single charge.

- Decreased the sensitivity of the on/off threshold, so that the lights installed closest to the high-mast lights (i.e., where it is brightest) will still turn on/off at the appropriate time. The on/off mechanism is determined by how much voltage is being received from the solar panel.

- Used a fixed arm support for the solar panel to hold it secure at the appropriate angle. This ensures that the solar panel is operating at optimal efficiency and makes the solar panel more stable and more difficult to tamper with.

- Used security screws for all fixings on the light to make it more difficult to tamper with.

- Laser-printed a logo on the light along with the following text “Light is the Property of [Settlement name redacted for privacy reasons], Ward [Redacted]. Not for Resale.” The idea to include the logo was inspired by the City of Cape Town, which also marks infrastructure that is easy to steal (e.g., water taps) to make it easy to identify. The logo also signals that the infrastructure is owned by the community and is being monitored.

2020-10-01

2021-03-31

Our primary outcomes of interest are:

1) light brightness (lux) to measure efficacy of the technology itself,

2) nighttime mobility as an indicator of willingness to be in public space at night,

3) nighttime activity as an indicator of willingness to be in public space at night,

4) satisfaction with light infrastructure, and

5) theft, vandalism, and technical malfunction (only of treatment group).

Light brightness (i.e., treatment intensity) is measured using a lux meter. Nighttime mobility will be measured in two ways. First, we will use novel pedestrian motion sensors to gather data about nighttime mobility by recording a trigger count (caused by passersby) every five minutes. Nighttime activity and light satisfaction will be measured with a survey. Theft, vandalism and technical malfunction will be measured by documenting service requests documented by the local maintenance team, as well as through an endline survey. The construction of each outcome variable is described below.

1) Average Lux: This outcome will be constructed by first taking the average of the minimum lux measurement and maximum lux measurement collected by the field team at each front door throughout the informal settlement. A path segment average is calculated by taking the average of all the average measurements calculated for each structure with a front door facing the path. For path segments with very few front doors facing onto the path segment, we pre-specified a set of points on the map to take an additional measurement to ensure that the Average Lux measurement for the path would not be biased. Depending on the length of the path segment, we took between three and eight measurements per path segment.

2) Average Motion: We will look at average motion at several levels of aggregation: a) average count per five-minute period, b) average count per hour at a given hour of the day, c) average number of pedestrians per day over all days in the study. We use pedestrian motion sensors developed by Sensen, a company that develops dataloggers for international development projects. These small devices are installed on structures throughout the informal settlement. They use proximity infrared (PIR) sensors to detect the presence of a person.

3) Nighttime Activity Index: Using an endline survey we will ask a battery of questions about nighttime activities, including whether residents use shared toilets at night, visit the local store (locally called a Spaza shop) at night, whether they go to church at night, or engage in a variety of other social activities. Each response will be coded as 1 for yes and 0 for no. All the responses will be summed for each participant to create a count index of nighttime activity.

4) and 5) Infrastructure Satisfaction Index: At the end of the endline survey we will ask a battery of questions about what residents in the informal settlement think of the public lighting situation, including whether the solar lights are placed in the right locations, whether the lights are bright enough (or too bright), and whether they would recommend the light to residents of another informal settlement. These will comprise yes-no questions and questions on a Likert scale. All the responses will be summed for each participant to create a count index of satisfaction with infrastructure.

We collect service request information using a WhatsApp hotline as well as directly from the trained, local maintenance team. The following information is recorded in the service request database: structure identification number, date of the report, problem type, problem notes, fix/response notes, status of the service request, date problem is addressed, and staff member that reports the problem.

Our secondary outcomes of interest are: 1) perceptions of safety, and 2) experience of crime, both of which will be measured through an endline survey. We treat these outcomes as secondary because, as can be seen from the bulk of the literature, it is very difficult to get an unbiased measure from self-report survey responses to questions about how people perceive their own safety or level of fear as well as their past experience of crime.

1) Perception of Safety Index: During the endline survey we will ask a battery of questions about how residents perceive their own safety and security in the informal settlement, including whether respondents feel safe in the informal settlement at night, whether they feel safe in the informal settlement during the day, and whether they like living in the specific informal settlement. Each response will be coded on a Likert scale. All the responses will be summed for each participant to create a perception of safety count index.

2) Experience of Crime Index: During the endline survey we will ask respondents if they or someone in their household has been the victim of a set of outdoor crimes (e.g., robbery, physical assault, house vandalism) in the preceding six months of the intervention (1st of October 2020 until 31st of March 2021). Each response will be coded as 1 for respondents who say they have not been a victim of a given crime (e.g., positive response) and 0 for respondents who indicate that they have been the victim of a given crime. All the responses will be summed for each participant to create a count index of experience of crime at night.

We use a cluster-randomized controlled trial to study the impact of an alternative public lighting technology that is intended to provide brighter lighting on the dense, narrow paths in informal settlements than is currently provided by the two existing high-mast lights. High-mast lights can provide bright lighting on paths that are nearby, but they can also cast strong shadows and provide insufficient lighting in areas that are farther away. Until now, the impact of public lighting has rarely been quantitatively evaluated in poor, urban neighborhoods. The randomization allows us to test both the efficacy of the new technology and the impact of a public lighting intervention.

We stratify the entire path network of one poor informal settlement in Cape Town, South Africa with about 800 houses into pedestrian path segments and compounds (cul-de-sacs or shared semi-private spaces within the settlement that are used by several households, but are often gated).

We randomize 114 path segments (out of 133 possible segments) of the informal settlement into 49 treatment and 65 control path segments and the 50 compounds into 24 treatment and 26 control compounds. By randomizing at the path segment and compound level, we ensure that the treatment makes logical sense to a pedestrian or other user of public space at night. In other words, the intervention results in lit path segments, rather than randomly lit houses that might create patchy non-uniform lighting, or disconnected lit areas that do not enable residents to pass from one part of the neighborhood to another on a completely lit route. In this way, we are able to compare lit paths and compounds to unlit paths and compounds to analyze the effects of living on a lit path segment (or compound) on a variety of outcomes. Moreover, we can test whether lit paths or compounds are used more often using data gathered with the pedestrian motion sensors.

On treatment path segments and compounds, all households with a front door facing onto to the path or compound receive a free light for the entire six-month study period, which they can keep afterwards. The control households will receive a free light at the end of the study.

A house is only considered to be part of a path segment or compound if the front door is facing it. Note that we initially identified 133 path segments, but only allocated a total of 114 (49 + 65) path segments to the treatment and control group. The reason is that some of these paths (5 of the treatment, 12 of the control group) did not have any houses with front doors on them. Hence, these paths are excluded from the experiment. The remaining two segments are the central streets, which we also removed from the experiment because they are vehicular through-routes and cannot be accurately measured with sensors. In addition, some houses do not have front doors that open onto a path segment or a compound (e.g., houses that face the central street) so these houses will be excluded from the analysis.

For compounds, randomization was done in the office by a computer.

For path segments, the randomization procedure was done slightly differently in order to ensure that not only scattered path segments were lit, but rather that routes would be lit that a person could logically walk and to make sure that we would cover the entire settlement. We implemented the following protocol to quasi-randomly select paths.

The informal settlement is split by two central streets on which cars can pass (one from North-South and one from East to West) and is also surrounded by formal (paved) vehicular roads. None of these larger roads are assigned to either the treatment or control group. Beginning in the North-West of the settlement, we assigned (roughly) every second pedestrian path/route from North to South and from East to West that connects a bordering road with a central street within the settlement to the treatment (i.e., all of the segments that make up the route are in the treatment group). Figure 1: Path Network Treatment Map, included in the Supplemental Information, should help make the procedure clear.

The location of a structure’s front door determines whether it is considered part of a treatment or control path/compound. All structures with a front door on a treatment path segment or compound received a light.

The unit of randomization is either the path segment or the compound. Individual households only receive the treatment if their front door opens onto a path segment or compound selected into the treatment group.

Yes

114 paths
50 compounds

For all household-level data, we expect to have the following number of observations: - 465 households on 114 path segments (200 treatment; 264 control). - 171 households on 50 compounds (86 treatment; 85 control). For the sensor data, each sensor will produce an observation for every five-minute period of every day for about one to two months (depending on battery life and vandalism). Thus, we will have 288 periods per day per sensor for every day that the sensors are functional. We have installed a total of 85 sensors on path segments throughout the settlement.

Paths
49 treatment paths with lights for households
65 control paths without lights for households

Compounds
24 treatment compounds with lights for households
26 control compounds without lights for households

For all outcomes, we use a beta (power) of 0.8 and an alpha (significance level) of 0.01. Our calculations are done for a two-level cluster randomized trial with path/compounds being the clusters. For all outcomes we use 3.75 as the average number of households per path or compound. Standard deviations for outcome variables were taken from our baseline survey in 2019. We present the primary outcomes for paths and compounds here. For conservative estimations we did not yet take into account that we have baseline values for our outcome variables. As these will decrease the unexplained variance across paths and compounds, our minimum detectable effect sizes are likely overestimated, i.e., we will have smaller minimum detectable effect sizes. Calculations were done using the 3ie Sample Size and Minimum Detectable Effect Calculator. Average Lux: We are powered to detect a minimum effect of 2.65 lux (std. dev = 6.35, mean at baseline = 1.85). 5-minute Path Motion: We are powered to detect a minimum effect of 2.5 triggers per 5-minute period (std. dev = 3.86, mean at baseline = 1.94). Nighttime Activity Index: We are powered to detect a minimum effect of 0.685 (std. dev = 1.41, mean at baseline = 3.15).

Many informal settlements have little or no public lighting. Few studies analyze the impact of streetlighting on life at night in these neighborhoods, though theory suggests it may help residents feel safe, spend more time outside, and access shared infrastructure after dark. We apply a cluster-randomized controlled trial to test the
efficacy and impact of solar lights mounted onto houses as an alternative to standard streetlights in one informal settlement in Cape Town. Treated areas were between 6 and 8 times brighter than control areas, partly because theft and vandalism were minimal. The treatment is also linked to increases in perceived safety, but not to changes in most nighttime activities or experience of crime. However, residents were more likely to use shared sanitation at night. This study is the first, to our knowledge, to quantify the impact of streetlighting in informal settlements and to study a distributed public lighting alternative.

Borofsky, Y. and Günther, I., Bringing Light To The Dark — Can Solar Public Lighting Improve Nighttime Life For The Urban Poor? Working Paper. 2022.

https://ethz.ch/content/dam/ethz/special-interest/gess/development-economics-dam/Publications/Light_Solar_Public.pdf