Using segmented regression to estimate the optimal illuminance for pedestrian reassurance

Abstract

CIE 115:2010 gives guidance for the design of road lighting that feeds into the design standards of many nations. There is, however, an emerging awareness that the guidance in CIE 115:2010 needs amendment, and this is being targeted through rapid and long-term revisions. One aim of the rapid revision is to provide an empirical basis for the guidance, with that to be founded in what is already known rather than waiting for further research to be conducted. For P-class lighting, applied to situations where pedestrians are the target user, it has been suggested that pedestrian reassurance (the feeling of safety) is a suitable basis for establishing optimal lighting conditions. This article analyses the relationship between measures of pedestrian reassurance (after-dark evaluations and the difference between daylight and after-dark evaluations) and the conventional measures of lighting in the P class (mean, minimum and uniformity of horizontal illuminance) using the results from previous field studies investigating this. For these data, the day-dark difference provided a better association with illuminance than did after-dark ratings. For the day-dark difference, the association with mean illuminance was better than that with minimum illuminance or uniformity. Segmented regression suggested an optimum mean illuminance of 8.76 lx. To fit within the existing P classes, this could be rounded down to class P3 (mean illuminance = 7.5 lx).

1. Introduction

CIE 115:2010 Lighting of Roads for Motor and Pedestrian Traffic¹ reports ‘current knowledge and experience’ about road lighting, including two series of lighting classes for motorists (M class) and pedestrians (P class). For those classes, it recommends lighting conditions, and it also describes a series of factors that can be used for choosing an appropriate class. CIE 115:2010 is an important document because those recommendations are the basis of national guidance for road lighting. For example, it is the basis of class selection in EN 13201-1:2014 and lighting conditions in EN 13201-2:2015,^2,3 which in turn are the basis of lighting design criteria in BS5489-1:2020.⁴

There are, however, some limitations to the guidance given in CIE 115:2010.^5,6 One key issue is that the document does not reveal the evidence used to determine the recommended lighting conditions nor the reasons for the factors and weightings used to establish a certain lighting class. This need is recognised by CIE, and hence a revision of CIE115:2010 is a priority focus of activity in Division 4 of the CIE. In the first place, this is intended to be a rapid revision, using evidence already available rather than waiting for further research to be funded and completed.

The current article aims to establish an empirical basis for optimal lighting conditions in the P class and does so through an analysis of the results of previous studies investigating pedestrian reassurance. Pedestrian reassurance describes ‘the confidence a pedestrian might gain from road lighting (and other factors) to walk along a footpath or road, in particular if walking alone after dark’,⁷ and is intended to encompass the terms used in past studies such as perceived safety,⁸ perceived danger^9,10 and fear of crime.¹¹ Reassurance is a consideration of social importance because people are more likely to walk when they feel safe,^12,13 particularly so for women.¹⁴ This is recognised in design guidance, with one document⁴ stating that a purpose of road lighting in subsidiary roads is to ‘allow pedestrians to … feel more secure [and] … helping to reduce fear of crime’, while a second document¹ states that good lighting in a residential area ‘imparts a greater sense of security’.

Reassurance is a useful consideration for establishing an optimal illuminance for P-class lighting. It requires a higher illuminance than that needed for the detection of trip hazards,^7,15 itself an important consideration because falls on public footpaths are a significant problem in terms of the number of cases, the severity of the resulting injury and the national cost.¹⁶ Another visual need of pedestrians is to evaluate the intentions of other people, but there is insufficient evidence as to what lighting conditions are desirable. For example, previous studies have tended to ask test participants to identify the identity of celebrities, typically of light skin tone, with a full-face view, and with an unlimited observation period, and these conditions do not represent experience.^17–20 Furthermore, sufficient light to appraise another person may be a source of glare on that other person, and this balance remains unresolved.

Road lighting is expected to support reassurance because it aids the visual evaluations of prospect, refuge and escape,²¹ and indeed road lighting is associated with locations considered safe to walk after dark.^22,23 That suggests road lighting should be installed where the need to support reassurance is pertinent, but it does not say how much light should be provided.

To estimate how a change in illuminance affects reassurance, previous field studies have tended to evaluate reassurance only after dark, with a small sample of locations of different illuminance, and have tended to conclude that the higher the illuminance received, the higher rating of reassurance. Such results are described in CIE 236:2019 as trivial because ‘no matter what light levels are used, the higher will be rated as safer, leading to a recommendation for that higher light level’.⁷ This stimulus range effect leads towards ever higher illuminances rather than establishing an optimal illuminance.²⁴

There is evidence that an optimal light level can be established by either of two changes to the standard approach: (1) by using a much larger sample of locations,²⁵ or (2) by conducting reassurance evaluations also in daylight and using the day-dark difference as the measure of lighting effectiveness – good lighting is that which minimises the difference between daylight and after-dark evaluations of the same location.²⁶

The aim of this article is to make an estimate of the optimal illuminance for pedestrian reassurance using results from the day-dark approach to measuring the effectiveness of lighting. By including all known studies using this method, it expands the sample size. The current article extends interim publication of the work²⁷ through the inclusion of additional studies, the correction of errors (incorrect conversion of the 5-point response scale used by Wei et al.²⁸ to a common 6-point scale, and incorrect reporting of samples in previous studies), analysis of optimal illuminance using after-dark evaluations alone, and the use of an independent statistician to apply segmented regression.

2. Method

2.1 Included data

The primary focus of analysis was studies of pedestrian reassurance under different lighting conditions using the day-dark method. Studies using only after-dark evaluations were also included to test the assumption that the day-dark difference provides better characterisation of reassurance. The inclusion criteria were:

A field study, i.e., in situ evaluations of locations.

A repeated measures design was used, in which each test participant evaluated all locations (in both daylight and after-dark, where the day-dark difference was analysed).

Five or more locations were evaluated.

Illuminance characteristics were derived from spot illuminances measured across an array of nodes, such as the approach defined in EN 13201-3:2015,²⁹ rather than a single spot measurement at a location.

Table 1 lists the studies that did not meet these criteria and were hence omitted from the analysis.

Table 1

Studies of pedestrian reassurance were omitted from the current analysis

Reason for exclusion	Excluded studies
Evaluations were not field studies, i.e., used instead photographs, VR environments or scale models	Boomsma and Stegg,³⁰ Bullough et al.,³¹ Loewen et al.,³² Wu and Kim³³
The evaluations were independent samples	Mattoni et al.,³⁴ Peña-García et al.³⁵
Five or fewer locations were evaluated	Akashi et al.,³⁶ Nair et al.,³⁷ Painter,³⁸ Rea et al.,³⁹ Vrij and Winkel⁴⁰
Illuminance was neither measured nor reported	Atkins et al.,¹¹ Blobaum and Hunecke,⁹ Nair et al.,⁴¹ Herbert and Davidson⁴²
Reported spot illuminance rather than measurement following EN 13201-3:2015	Portnov et al.,²⁵ Svechkina et al.⁴³
Reported the change in ratings under different lighting conditions, but not the absolute ratings of each condition	Knight⁸

The minimum threshold of five locations was set as an attempt to offset the expected bias from the stimulus range effect. While this was an arbitrary threshold, exclusion of the five studies not meeting this threshold (see Table 1) removed 11 cases from the after-dark analysis but none from the day-dark analysis. The five omitted studies reported evaluations of only two or three locations (or the same location with two or three lighting conditions).

Table 2 shows those studies that met the inclusion criteria and were therefore included in the analysis. Five studies^28,44–47 examined pedestrian reassurance in urban roads, one study examined reassurance in parking lots,²⁶ and one study was located in an arboretum with lighting from low level bollards rather than lamp posts.⁴⁸

Table 2

Studies of pedestrian reassurance included in the current analysis show the locations of illuminance and evaluation data in the original publications. All studies used the day-dark approach except Himschoot et al.⁴⁸ which sought only after-dark evaluations

Study	Location	No. of locations	Lamp type	No. of test participants	Reassurance evaluation (a)	Response scale range	Location of data used in the current analysis
Study	Location	No. of locations	Lamp type	No. of test participants	Reassurance evaluation (a)	Response scale range	Illuminance	After-dark ratings	Day-dark differences
Fotios et al.⁴⁴ (b)	Sheffield, UK	8	Mixed	24	Composite	1–6	Table 1	Supplemental data	Table 5
Gorjimahlabani⁴⁵	Sheffield, UK	12	LED	122	Composite	1–6	Table 5.2	Table 6.16	Table 6.16
Liachenko Monteiro⁴⁶ (c)	Sheffield, UK	16	Mixed	34	Composite	1–6	Table 21	Appendix H	Table 25
Unwin⁴⁷ (d)	Sheffield, UK	9	Mixed	46	Single	1–6	Table 6.1	Figure 6.10	Table 6.12
Wei et al.²⁸	Ljubljana, Slovenia	11	Mixed	35	Composite	1–5	Table 1	Table 4	Table 7
Boyce et al.²⁶ (e)	Albany, NY	24	Not reported	18	Single	1–7	Figure 7	Not reported	Figure 7
Himschoot et al.⁴⁸	An arboretum in the United States (f)	8	LED	156	Single	5 pt (g)	Table 2	Table 2	Not done

(a)

The evaluations focused on perceived safety at that location: composite means factor analysis was used to derive a single measure from multiple ratings scales; single means only one relevant scale was used.

(b)

This is Field Study 1 in Liachenko Monteiro⁴⁶

(c)

This is the work labelled as Field study 2.

(d)

This is the work labelled as ‘Winter study’.

(e)

This is the work labelled as Field study 3.

(f)

This location was described as ‘a large U.S. university-managed arboretum within a multi-mile wooded trail system’.

(g)

See comment on this scale in section 2.2.2.

All studies were included where possible in both analyses (the day-dark difference and the after-dark ratings), with the reassurance data being compared with mean, minimum and uniformity of illuminance. Himschoot et al.⁴⁸ recorded evaluations only after dark, so it is omitted from the day-dark difference analysis. Boyce et al.²⁶ recorded evaluations in both daylight and after dark, thus to determine the day-dark difference, but did not report the after-dark evaluations, and were therefore omitted from the after-dark analysis.

Three studies^44–46 reported a composite score of reassurance determined using factor analysis of day-dark differences determined from the responses to multiple questions, including How safe do you think this street is?, How anxious do you feel when walking down this street? and I would rather avoid this street if I could. Similarly, Wei et al.²⁸ reported the composite score determined from three questions addressing safety, risk and comfort. However, only one study⁴⁵ followed that same process for analysis of the after-dark ratings. For the other three studies, the current analysis used:

Fotios et al.: The mean of responses to the three questions about safety, anxiousness and avoidance.⁴⁴

Liachenko Monteiro: The mean of responses to the two questions about safety and anxiousness. Responses to the avoidance question were omitted since it is not clear whether or not the scale direction had been reversed in the reported data.⁴⁶

Wei et al.: Responses to only the question about safety. Responses to the question about risk were omitted since it is not clear whether or not the scale direction had been reversed in the reported data.²⁸

Boyce et al.²⁶ asked multiple questions but reported only the responses to the question asking how risky it would be to walk alone at night. Finally, two studies asked only one question, that being about the feeling of safety.^47,48

2.2 Data manipulation

Consideration was given to the inclusion of individual locations in the original studies, the rating scales used to record evaluations, and the reported lighting conditions.

2.2.1 Location omission

Data for all of the locations reported in the previous studies were retained except for three. The two locations in Fotios et al.⁴⁴ identified as being different types of environments were omitted to follow the collection labelled as ‘N = 8’ in that study. Location R8 in Wei et al.²⁸ was omitted as it was unlit.

2.2.2 Response scale conversion

The analysis reported here assumes that the reassurance questionnaires used a 6-point rating scale where 1 represents low reassurance and 6 represents high reassurance. Such a scale was originally used in four studies.^44–47 For some questions, the direction of the response scale was reversed (i.e. 1 represented high reassurance and 6 represented low reassurance), this being a step to check for attentive responding. Where the original study reported the reversed data, this was used in the current analysis; where it is clear that the ratings have not been reversed to account for scale direction, this was done for the current analysis; where there is uncertainty, these data were omitted from the analysis.

Wei et al.²⁸ used a 5-point rating scale ranging from 1 to 5. These responses were converted to a 1–6 response scale using Equation (1), where R_1–5 is the original rating using a 1–5 response scale and R_1–6 is the rating when converted to a 1–6 response scale.

R_{1 - 6} = ((R_{1 - 5} - 1) 5 / 4) + 1

(1)

Day-dark differences determined from a 1 to 6 rating scale would range from 5 (a day rating of 6 and a dark rating of 1) to 0 (the same rating for daylight and after dark) or even to negative values (the daylight rating being lower than the dark rating). The day-dark differences reported by Wei et al. were therefore multiplied by 5/4 to convert them to a day-dark difference range of 0–5. Boyce et al.²⁶ used a 1–7 response range, meaning that their day-dark difference range of 0–6 required multiplication by 5/6 to convert it to that which would be gained from ratings using a 1–6 response scale.

The studies listed in Table 2 tended to assign each category in their response scale the integer 1, 2, … 6 and reported the average value of these ratings. Himschoot et al.⁴⁸ instead reported the proportion of responses awarded to each of their five categories. For the current analysis, these categories were assumed to be represented by the integers 1 to 5, a mean rating was estimated using those proportions as a weighting for the assigned integer, and the resulting mean converted to that for a 1–6 response range using Equation (1).

2.2.3 Illuminance data

The current analysis investigated the association between reassurance (the day-dark difference and the after-dark ratings) and the mean, minimum and uniformity of illuminance, these being the primary characteristics of design guidance for P class lighting.^1,4 In three studies^28,45,46 all three values were reported. In two studies^44,47 the mean illuminance was reported but with only the minimum or uniformity: the third value was thus calculated from the other two values. Note that for the current analysis study, uniformity of illuminance was defined as the ratio of minimum illuminance to mean illuminance. In the final two studies^26,48 only mean illuminance of each location was reported, so it was not possible to determine the minimum illuminances or uniformities. Note that Boyce et al. reported median illuminance rather than mean. The current analysis, therefore, used their median illuminance, following a study that suggested that mean and median illuminance are sufficiently similar.⁴⁹ Table 3 shows the studies included in each specific analysis.

Table 3

Inclusion of studies within the analysis of the prediction of reassurance by the given lighting measure, according to the availability of data

Study	After dark			Day-dark difference
Study	Mean	Minimum	Uniformity	Mean	Minimum	Uniformity
Fotios et al.⁴⁴	✓	✓	✓	✓	✓	✓
Gorjimahlabani⁴⁵	✓	✓	✓	✓	✓	✓
Liachenko Monteiro⁴⁶	✓	✓	✓	✓	✓	✓
Unwin⁴⁷	✓	✓	✓	✓	✓	✓
Wei et al.²⁸	✓	✓	✓	✓	✓	✓
Boyce et al.²⁶				✓
Himschoot et al.⁴⁸	✓

2.3 Data analysis

Regression analysis was used to explore the relationship between the measures of lighting (mean, minimum and uniformity of illuminance) and evaluations of reassurance (after-dark ratings and the day-dark difference). A lighting measure providing a sufficient prediction of reassurance was one for which the best-fit line suggested a statistically significant relationship (p < 0.05) and for which the plot of residuals suggested a random distribution. The lighting measure(s) suggesting the better prediction of reassurance were those providing the higher coefficient of determination (R²) – the higher the R² value, the greater the percentage of change in the reassurance ratings that is explained by the change in that specific lighting measure.

For the lighting measure(s) suggesting good prediction of reassurance, the optimal value of that measure was determined using segmented regression.^25,50,51 Segmented regression separates the data into two (or more) segments, separated at a break point, with each segment characterised by a separate linear equation. The break point between the two segments defines the optimal value of that lighting measure and for the current analysis was defined through iterative choices of the break point such that: (i) the region above the break point is a flat line (a slope of 0), indicating that illuminances higher than the break point bring no further increase in reassurance; and (ii) the region below the break point displays a significant relationship between lighting measure and reassurance, with a lower value of (e.g.) mean illuminance suggesting a lower degree of reassurance.

Segmented regression was implemented using the segmented package in R. Segmented regression models were fitted to each dataset separately, applying either one or two break points (for two or three segments, respectively). The gradient of the final segment was constrained to zero (horizontal fitted line) where possible. Likelihood ratio tests (used for comparing the goodness of fit of two competing statistical models) were used to assess the improvement in fit for the segmented model versus a simple linear fit, and for two versus one break points.

3. Results

3.1 Regression analysis

The distributions considered are the reported mean reassurance rating (after-dark or day-dark difference) and the reported mean, minimum and uniformity of illuminance for each location. The regression considered linear and logarithmic best-fit lines, with a linear equation proposed unless the logarithmic equation produced a higher R² value. Table 4 shows the results of the regression analysis. To check the sensitivity of these relationships, the regressions were repeated with individual studies removed. The removal of any one study did not lead to a marked change in the relationships between the illuminance and reassurance variables. Figures 1 and 2 show these relationships.

Table 4

Results of regression analysis exploring lighting measures for predicting reassurance

Reassurance	Illuminance characteristic	Equation*		N	R ²	p value	Residuals
Day-dark difference	Mean	Log	R_DD = 2.1073 − 0.655 ln(E_mean)	80	0.42	<0.001	No trend
	Minimum	Log	R_DD = 0.6138 − 0.279 ln(E_min)	56	0.44	<0.001	No trend
	Uniformity	Linear	R_DD = 0.801 − 0.9842 (U)	56	0.16	0.02	No trend
After dark	Mean	Linear	R_AD = 3.718 + 0.0559 (E_mean)	64	0.10	0.009	No trend
	Minimum	Log	R_AD = 4.0204 + 0.3544 ln(E_min)	56	0.30	<0.001	No trend
	Uniformity	Linear	R_AD = 3.6212 + 1.9142 (U)	56	0.27	<0.001	No trend

Reassurance evaluations are noted as R_DD for the day-dark difference and R_AD for the after-dark rating.

Figure 1

The day-dark difference in reassurance plotted against mean illuminance (top), minimum illuminance (middle) and uniformity of illuminance (bottom). In these graphs, a day-dark difference approaching 0 indicates good lighting – the difference between daylight and after-dark evaluations is minimised

Figure 2

The after-dark rating of reassurance plotted against mean illuminance (top), minimum illuminance (middle) and uniformity of illuminance (bottom). In these graphs, a higher rating indicates a higher feeling of safety

In all cases, the association between lighting and reassurance is suggested to be statistically significant (p < 0.05), and no non-random patterns were observed in the residual plots. For the day-dark difference, mean and minimum illuminance are better predictors than uniformity; for the after-dark ratings, minimum illuminance and uniformity are better predictors than mean illuminance. With these data, the weakest association found is that between after-dark ratings of reassurance and mean illuminance. There is a stronger association between illuminance and reassurance for the day-dark difference than for after-dark ratings. Overall, the data presented in Table 4 suggest that using mean or minimum illuminance to predict the day-dark difference is the best approach for predicting the benefit of lighting for pedestrian reassurance.

3.2 Segmented regression

Segmented regression revealed break points for the regression of the day-dark difference against either mean illuminance or minimum illuminance (Figure 3). These suggest optimal mean and minimum illuminances of 8.76 lx (95% confidence interval 6.20 lx to 11.32 lx) and 1.54 lx (0.88 lx to 2.20 lx) respectively. For the other four data sets, segmented regression was not able to reveal a break point leading to a significant improvement in fit.

Figure 3

Segmented regression lines fitted to the day-dark difference and mean illuminance (top) or minimum illuminance (bottom)

4. Discussion

4.1 Main findings

To support the rapid revision of CIE 115:2010,¹ this article reports an analysis of the results from previous field studies about the relationship between lighting conditions and pedestrian reassurance in order to estimate the optimal illuminance for pedestrian reassurance. Six combinations of three lighting criteria (mean, minimum and uniformity of illuminance) and two reassurance evaluations (after-dark and the day-dark difference) were examined. It was found that a change in mean or minimum illuminance exhibits a better association with the day-dark difference in reassurance than with after-dark ratings (Table 4). Segmented regression was able to identify optimal values (break points) in the regressions of day-dark difference against either mean or minimum illuminance but was not able to reveal such points in any of the other four combinations.

The rapid revision of CIE 115:2010 intends that the existing set of lighting classes is retained. The optimal criteria determined in the current analysis (mean illuminance = 8.76 lx, minimum illuminance = 1.54 lx) lie between classes P3 (mean = 7.5 lx, minimum = 1.5 lx) and P2 (mean = 10 lx, minimum = 2.0 lx). On the basis that the lower illuminance of P3 is still within the confidence interval of the estimated optimal values, and that lower illuminance contributes to a reduction in the unwanted impacts of lighting after dark, then for the locations included in the current analysis, it can be concluded that lighting class P3 would be the optimal choice.

Figure 3 shows that the horizontal segment in each graph lies at a day-dark difference of about 0.5, close to the day-dark difference of 0 obtained when the same degree of reassurance is reported in daylight and after dark. This indicates what we should expect from lighting after dark – it has the potential to raise reassurance towards that found in daylight but is unlikely to reach the same level. That does not necessarily mean a high level of reassurance at a location, because other factors, such as signs of incivility, may result in a low level of reassurance even in daylight; instead, it means this is the best that road lighting is able to do.

4.2 Reasons to use different illuminances

Changes in context might suggest different illuminances are optimal. For example, in Figure 1 (top), which plots the day-dark difference against mean illuminance, the data points from the Boyce et al. study tend to lie above the best-fit line, indicating a greater day-dark difference at that illuminance. This suggests that parking lots, the locations of that study, would benefit from higher illuminance. In Figure 2 (top), which plots after-dark reassurance against mean illuminance, the data points from Himschoot et al. tend to lie above the best-fit line, indicating a greater feeling of safety: this suggests that in that location, a lower illuminance could be used.

These differences may be associated with the opportunities for concealment and entrapment. Concealment refers to hiding spots for potential offenders, such as bushes, walls and shadows; entrapment refers to physical features that impose a barrier to escape, and these will differ between open spaces and car parks. Narrow alleys are less likely to suggest good possibilities for escape and are therefore considered more dangerous than wide alleys.⁵² Blöbaum and Hunecke⁹ found in their field study that increasing the light level in areas of low entrapment led to a significant increase in reassurance, but in areas of high entrapment, the increased light level had relatively little effect.

For some situations, there is evidence to suggest that a higher light level would be beneficial.

The detection of pavement hazards is hindered by the need to (or expectation to) conduct a face evaluation task in parallel, but this can be offset by a higher light level.⁵³ If the same impact of multi-tasking can be assumed for other visual tasks of pedestrians, then a higher light level would be beneficial in situations with higher cognitive demand, such as higher numbers of pedestrians, mixed-use routes, higher numbers of vehicles and more-hazardous pavement surfaces.

Higher light levels can reduce the reaction time to detection of a target^54–56 and may therefore offset the increased risk of a road traffic collision after dark on roads with a higher speed limit.^57–60 However, while higher luminances are associated with decreased crash risk, including decreased risk of crashes involving a pedestrian, as was reported by Jacket and Frith,⁶¹ it remains to be established whether higher luminances mitigate the increased risk of higher speeds.

It has been shown that darkness deters walking and cycling.^62–65 For cycling, it has been shown that the deterrent effect of darkness is reduced at higher light levels⁶⁶ and thus might be used as a means of promoting active travel, but that remains to be demonstrated for pedestrians.

In contrast, lower levels of road lighting may be suitable where the speed limit is less than 20 mph, where the footpath poses little likelihood of a trip hazard, in pedestrian-only footpaths, and where there are few other pedestrians or traffic.

Finally, while the weighting system of CIE 115:2010 assumes a cumulative effect for different weighting factors, this remains to be proven. It may be the case that changing the lighting class by one step may be sufficient mitigation for two (or more) risk factors appearing simultaneously, which would otherwise each prompt one step change, rather than needing to change the lighting class by two steps.

4.3 Limitations

Subjective measures of reassurance are prone to socially desirable responding⁶⁷ and other response biases.⁶⁸ Furthermore, as stated preferences, the extent to which they impact behaviour (the revealed preference) is unknown. While alternative approaches have been explored^65,69,70 there are insufficient data available with these methods to demonstrate reproducibility and to identify optimal illuminance.

The data included in this analysis are limited by availability. For the analysis of day-dark difference and mean illuminance, the 80 overall locations are those reported in six studies, with those individual studies reporting evaluations of 8 to 24 locations by 18 to 122 test participants (Table 2). Furthermore, 45 of those locations were in one city. It is suggested, however, that these data are suitable for the required rapid revision of CIE115:2010 for two reasons: (i) the data sources are revealed, meaning that others can test by replication the conclusions drawn and question the context relevance; and (ii) the sample of locations and observers is larger than those included in Simons et al.⁷¹ (12 locations evaluated by 13 people), which was considered sufficient evidence for the light levels recommended in BS5489-3:1992.⁷²

With these data, mean illuminance explains 42% of the variance in reassurance after dark (Table 4). Some of that variance can be ascribed to uncertainty in the measurements of illuminance and reassurance in the underlying studies – neither is easy to measure with precision. Further sources of variance may be associated with physical features of the environment, such as route width and the presence and proximity of greenery.^52,73,74 Understanding the interaction between lighting and these other variables will help to identify where a change in light level is suitable.

5. Conclusions

Data from several studies of pedestrian reassurance were collated and analysed to determine an optimal lighting condition for P-class roads. It was found that using mean or minimum illuminance to predict the day-dark difference resulted in better relationships, these offering the highest R² values and were the only two cases for which segmented regression revealed a break point. Segmented regression suggested optimal illuminances of 8.76 lx (mean illuminance) and 1.54 lx (minimum illuminance). They are considered optimal because lower illuminances are associated with a decrease in reassurance, but higher illuminances bring no further increases in reassurance. To keep within the lighting classes of CIE115:2010, this suggests lighting class P3 (mean = 7.5 lx, minimum = 1.5 lx) is optimal for pedestrian reassurance. Some situations are suggested where higher or lower illuminances might be suitable, but those changes remain to be verified.

Footnotes

Acknowledgements

The segmented regression analysis was conducted by Pete Laud, Statistical Services Unit, University of Sheffield.

Correction (April 2026):

Reference 23 of this article has been updated since its original publication.

Declaration of conflicting interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

S Fotios

References

Commission Internationale de l’Éclairage. Lighting of Roads for Motor and Pedestrian Traffic. CIE 115:2010. Vienna: CIE, 2010

Commission Internationale de l’Éclairage. Road Lighting. Part 1: Guidelines on Selection of Lighting Classes. CEN/TR 13201-1:2014. Brussels: CEN, 2014.

Commission Internationale de l’Éclairage. Road Lighting. Part 2: Performance Requirements. CEN/TR 13201-2:2015. Brussels: CEN, 2015.

British Standards Institution. Design of Road Lighting. Part 1: Lighting of Roads and Public Amenity Areas – Code of Practice. BS 5489-1:2020. London: BSI, 2020.

Fotios

Gibbons

. Road lighting research for drivers and pedestrians: The basis of luminance and illuminance recommendations. Lighting Research and Technology 2018; 50: 154–186.

Fotios

. A review of design recommendations for P-class road lighting in European and CIE documents – Part 1: Parameters for choosing a lighting class. Lighting Research and Technology 2020; 52: 607–625.

Commission Internationale de l’Éclairage. Lighting for Pedestrians: A Summary of Empirical Data. CIE 236:2019. CIE: Vienna. 2019.

Knight

. Field surveys investigating the effect of lamp spectrum on the perception of safety and comfort at night. Lighting Research and Technology 2010; 42: 313–330.

Blöbaum

Hunecke

. Perceived danger in urban public space: the impacts of physical features and personal factors. Environment and Behavior 2005; 37: 465–486.

10.

Johansson

Pedersen

Maleetipwan-Mattson

, et al. Perceived outdoor lighting quality (POLQ): a lighting assessment tool. Journal of Environmental Psychology 2014; 39: 14–21.

11.

Atkins

Husain

Storey

. The influence of street lighting on crime and fear of crime. Crime Prevention Unit, Paper No. 28. London: Home Office, 1981.

12.

Foster

Hooper

Knuiman

, et al. Safe RESIDential Environments? A longitudinal analysis of the influence of crime-related safety on walking. International Journal of Behavioral Nutrition and Physical Activity 2016; 13: 22–30.

13.

Mason

Kearns

Livingston

. ‘Safe going’: the influence of crime rates and perceived crime and safety on walking in deprived neighbourhoods. Social Science and Medicine 2013; 91: 15–24

14.

Foster

Hillsdon

Thorogood

. Environmental perceptions and walking in English adults. Journal of Epidemiology & Community Health 2004; 58: 924928

15.

Fotios

Cheal

. Obstacle detection: a pilot study investigating the effects of lamp type, illuminance and age. Lighting Research and Technology 2009; 41: 321–342.

16.

Fotios

Uttley

. Illuminance required to detect a pavement obstacle of critical size. Lighting Research and Technology 2018; 50: 390–404.

17.

Fotios

Johansson

. Appraising the intention of other people: ecological validity and procedures for investigating effects of lighting for pedestrians. Lighting Research and Technology 2019; 51: 111–130.

18.

Fotios

Mao

Hamoodh

, et al. Using relative visual performance to predict performance of an interpersonal evaluation task with variation in adaptation luminance, observer age, skin tone, pavement reflection and interpersonal distance. Lighting Research and Technology 2024; 56: 283–294.

19.

Fotios

Uttley

Fox

. Exploring the nature of visual fixations on other pedestrians. Lighting Research and Technology 2018; 50: 511–521.

20.

Mao

Fotios

. Road lighting for pedestrians: comparing full-face and 3/4 views in a facial emotion recognition task. Lighting Research and Technology 2025; 57: 250–263.

21.

Fisher

Nasar

. Fear of crime in relation to three exterior site features: prospect, refuge, and escape. Environment and Behavior 1992; 24: 35–65

22.

Fotios

Unwin

Farrall

. Road lighting and pedestrian reassurance after dark: a review. Lighting Research and Technology 2015; 47: 449–469.

23.

Park

Garcia

. Pedestrian safety perception and urban street settings. International Journal of Sustainable Transportation 2020; 14:11: 860–871.

24.

Fotios

Castleton

. Specifying enough light to feel reassured on pedestrian footpaths. Leukos 2016; 12; 235–243.

25.

Portnov

Fotios

Saad

, et al. Establishing optimal illuminance for pedestrian reassurance using segmented regression. Lighting Research and Technology 2024; 56: 260–273.

26.

Boyce

Eklund

Hamilton

, et al. Perceptions of safety at night in different lighting conditions. Lighting Research and Technology 2000; 32: 79–91.

27.

Fotios

Uttley

Gorjimahlabani

. Optimal illuminance for pedestrian reassurance. Proceedings of the CIE 2025 Midterm Session, Vienna, Austria, 7–9 July 2025: 299–305.

28.

Wei

Bizjak

Kobav

. Evaluating the impact of road lighting on pedestrian reassurance through the day-dark approach. Lighting Research and Technology 2024; 57: 264-277.

29.

Commission Internationale de l’Éclairage. Road Lighting. Part 3: Calculation of Performance. CEN/TR 13201-3:2015. Brussels: CEN, 2015.

30.

Boomsma

Steg

. Feeling safe in the dark: Examining the effect of entrapment, lighting levels, and gender on feelings of safety and lighting policy acceptability. Environment and Behavior 2014; 46: 193–212.

31.

Bullough

Snyder

Kiefer

. Impacts of average illuminance, spectral distribution, and uniformity on brightness and safety perceptions under parking lot lighting. Lighting Research and Technology 2019; 52: 626–640.

32.

Loewen

Steel

Suedfeld

. Perceived safety from crime in the urban environment. Journal of Environmental Psychology 1993; 13: 323–331.

33.

Kim

. The relationship between the pedestrian lighting environment and perceived safety. Journal of Digital Landscape Architecture 2016; 1: 57–66

34.

Mattoni

Burattini

Bisegna

, et al. The pedestrian’s perspective: how do illuminance variations affect reassurance? Proceedings of the 2017 IEEE International Conference on Environment and Electrical Engineering and 2017 IEEE Industrial and Commercial Power Systems Europe (EEEIC / I&CPS Europe), Milan, Italy, 6–9 June 2017.

35.

Peña-García

Hurtado

Aguilar-Luzón

. Impact of public lighting on pedestrians’ perception of safety and well-being. Safety Science 2015; 78: 142–148.

36.

Akashi

Rea

Morante

. Progress Report: Improving Acceptance and Use of Energy-Efficient Lighting. Unified Photometry: An Energy-Efficient Street Lighting Demonstration in Easthampton, Massachusetts. Troy, NY: Lighting Research Center, 2004.

37.

Nair

McNair

Ditton

. Street lighting: unexpected benefits to young pedestrians from improvement. Lighting Research and Technology 1997; 29(3); 143–149.

38.

Painter

. The impact of street lighting on crime, fear and pedestrian street use. Security Journal 1994; 5; 116–124.

39.

Rea

Bullough

Brons

. Parking lot lighting based upon predictions of scene brightness and personal safety. Lighting Research and Technology 2017; 49: 293–304.

40.

Vrij

Winkel

. Characteristics of the built environment and fear of crime: a research note on interventions in unsafe locations. Deviant Behavior 1991; 12: 203–215.

41.

Nair

Ditton

Phillips

. Environmental improvements and the fear of crime: the sad case of the ‘pond’ area in Glasgow. British Journal of Criminology 1993; 33: 555–561.

42.

Herbert

Davidson

. Modifying the built environment: the impact of improved street lighting. Geoforum 1994; 25: 339–350.

43.

Svechkina

Trop

Portnov

. How much lighting is required to feel safe when walking through the streets at night? Sustainability 2020; 12: 3133

44.

Fotios

Liachenko Monteiro

Uttley

. Evaluation of pedestrian reassurance gained by higher illuminances in residential streets using the day-dark approach. Lighting Research and Technology 2019; 51: 557–575.

45.

Gorjimahlabani

. Road lighting for pedestrian reassurance: an investigation of methods and optimal illuminance. PhD thesis. University of Sheffield, Sheffield, UK, 2025.

46.

Liachenko Monteiro

. Road lighting for pedestrian reassurance: consideration of methods and new metric. PhD thesis. University of Sheffield, Sheffield, UK, 2021.

47.

Unwin

. An investigation of the effect of street lighting on pedestrian reassurance in residential environments. PhD thesis. University of Sheffield, Sheffield, UK, 2014.

48.

Himschoot

Crump

Buckley

, et al. Feelings of safety for visitors recreating outdoors at night in different artificial lighting conditions. Journal of Environmental Psychology 2024; 97: 102374.

49.

Fotios

Robbins

. Research Note: Describing average illuminance for P-class roads. Lighting Research and Technology 2020; 52: 1057–1062.

50.

Muggeo

VMR

. Estimating regression models with unknown break-points. Statistics in Medicine 2003; 22: 3055–3071.

51.

Liu

Zidek

. On segmented multivariate regression. Statistica Sinica 1997; 7: 497–452.

52.

Herzog

Flynn-Smith

. Preference and perceived danger as a function of the perceived curvature, length, and width of urban alleys. Environment and Behavior 2001; 33: 653–666.

53.

Mao

Fotios

. Lighting for pedestrians: does multi-tasking affect the optimal light level determined using task performance? Lighting Research and Technology 2022; 54: 33–60.

54.

Rea

Bierman

, et al. Evaluating light source efficacy under mesopic conditions using reaction times. Journal of the Illuminating Engineering Society 1997; 26: 125–138.

55.

Wood

Isoardi

Black

, et al. Night-time driving visibility associated with LED streetlight dimming. Accident Analysis and Prevention 2018; 121: 295–300.

56.

Fotios

Cheal

Fox

, et al. The effect of fog on detection of driving hazards after dark. Lighting Research and Technology 2018; 50: 1024–1044.

57.

Gårder

. The impact of speed and other variables on pedestrian safety in Maine. Accident Analysis and Prevention 2004; 36: 533-542.

58.

Rosén

Sander

. Pedestrian fatality risk as a function of car impact speed. Accident Analysis and Prevention 2009; 41: 536-542.

59.

Tefft

. Impact speed and a pedestrian’s risk of severe injury or death. Accident Analysis and Prevention 2013; 50: 871–878.

60.

Elvik

Christensen

Amundsen

. Speed and road accidents: an evaluation of the power model. TOI Report 740/2004. Oslo: The Institute of Transport Economics (TOI), 2004.

61.

Jacket

Frith

. Quantifying the impact of road lighting on road safety: a New Zealand study. IATSS Research 2013; 36: 139–145.

62.

Uttley

Fotios

. Using the daylight savings clock change to show ambient light conditions significantly influence active travel. Journal of Environmental Psychology 2017; 53; 1–10.

63.

Fotios

Uttley

Fox

. A whole-year approach showing that ambient light level influences walking and cycling. Lighting Research and Technology 2019; 51: 55–64.

64.

Fotios

Robbins

. Effect of ambient light on the numbers of motorized vehicles, cyclists and pedestrians. Transportation Research Record 2022; 2676: 593–605.

65.

Fotios

Uttley

Yesiltepe

, et al. Variance in odds ratios for estimating the deterrent effect of darkness on cycling: variation due to the choice of case and control hours. PLoS One 2024; 19: e0311964.

66.

Uttley

Fotios

Lovelace

. Road lighting density and brightness linked with increased cycling rates after dark. PLoS One 2020; 15: e0233105.

67.

Sutton

Farrall

. Gender, socially desirable responding and the fear of crime. British Journal of Criminology 2005; 45: 212–224.

68.

Fotios

. Using category rating to evaluate the lit environment: is a meaningful opinion captured? Leukos 2019; 15: 127–142.

69.

Patching

Rahm

Jansson

, et al. A new method of random environmental walking for assessing behavioral preferences for different lighting applications. Frontiers in Psychology 2017; 8: 345.

70.

Jedon

Haans

de Kort

. Perceived environmental safety and alertness in virtual urban environments. Journal of Transport & Health 2025; 44: 102138.

71.

Simons

Hargroves

Pollard

, et al. Lightng Criteria for Residential Roads and Areas. Venice: CIE, 1987: 274–277.

72.

British Standards Institution. Road Lighting: Part 3; Code of Practice for Lighting for Subsidiary Roads and Associated Pedestrian Areas. BS5489-3:1992. London: BSI, 1992.

73.

Nikunen

Korpela

. The effects of scene contents and focus of light on perceived restorativeness, fear and preference in nightscapes. Journal of Environmental Planning and Management 2012; 55: 453–468.

74.

Zhao

Deng

Luo

, et al. Preferred streets: assessing the impact of the street environment on cycling behaviors using the geographically weighted regression. Transportation 2025; 52: 1485–1511.