Do photos help resolution of FixMyStreet reports?


FixMyStreet allows people to upload images along with a report. This can quickly provide the authority with more details of the issue than might be passed along in the written description, and lead to quicker evaluation and prioritisation of the repair. For problems that are hard to locate geographically by description (or where the pin has been dropped inaccurately), images might also help council staff locate and deal with the problem correctly.

In 2019, 35% of reports included photos. Accounting for several other possible factors,  reports with photos were around 15% more likely to be recorded as fixed than reports without a photo. In absolute terms, reports with photos were fixed at a rate two percentage points higher. This varies by category, with photos having a much stronger effect (highways enquiries and reports made in parks and open space) in some categories, and in other categories photos having a small negative effect in the resolution (reports of pavement issues and rights of way).

In general, these results suggest that attaching photos is not only useful for authorities, but can make it more likely that reporters have their problem resolved. There is a significant reservation that photos are much more useful for some kinds of reports than others. In terms of impacts on the service, when photos can convey useful information that helps lead to a resolution, users should be encouraged to attach them. Where photos are less helpful (such as problems encountered mostly at night), other prompt suggestions or asset selection tools may help lead to more repairs.


FixMyStreet has two ways of knowing if a report has been fixed: it can be reported fixed by a user, or by the authority. As user-reported fix rates are inconsistent (the status is more likely to be updated by superusers, who are also more likely to post photos), this analysis focuses only on reports where there is a reported fix by the council.

To select this data, the overall set of FixMyStreet reports was limited to reports in councils where there has been a report marked as fixed by the council rather than users, and then narrowed down to reports made after the first reported fix in that area. This identifies when a council starts feeding back information about repairs into FixMyStreet. Additional user-reported fixes were removed. This left 663,591 reports that were either not reported fixed, or were reported fixed by the council. As a result of this selection the ‘repair rates’ in this data are not representative of the repair rates in FixMyStreet in general (as some information about successful fixes has been discarded). This means the interesting results of the analysis are the different sizes of change rather than the absolute rate of fixes.

A logistic regression was used to examine the effect of having a photo on the chances a report had been marked fixed. To control for other possible drivers of the repair rate, also included in the model was the category of report, the cobrand (if the report was made through or integrated with a council website), and the year and month the report was made in. Different category names in different areas were grouped using the Sheffield B meta-categories for FixMyStreet.  Month was included as a set of 12 categorical variables to account for different reports and repair rates in different months for seasonable problems like potholes.

The McFadden pseudo R2 for this model suggests these factors collectively are accounting for around 15% of the variation in the repair rate.

In this model, having a photo leads to a 15% [95% confidence between 13.5, 16.6] increase in the probability of a fix. To put this in absolute terms, the probability of a report being reported fixed in the year 2019 was calculated (34%), and this was then adjusted by the effect of a photo to (37.17%). This means that if no reports had photos, and suddenly all reports had photos, the model would predict an increase in the repair rate by 3.2 percentage points [2.9, 3.5] across all reports.

To have a sense of how this effect varies by the type of report, the model was rerun looking at reports in each category in turn. While most categories of report that did not report an effect had a small number of cases, reports made about drainage and street furniture (which had around 35,000 and 20,000 reports respectively) did not show any effect on repair of a photo being submitted. For some categories there were effects larger than the effect in the dataset as a whole. For others there was a negative effect, where including a photo appeared to lead to a reduced rate of fixes.

Table 1 shows the categories of report for which there was a significant effect of including a photo in the report, controlling for cobrand and time of year. This table shows two numbers for each category, the relative increase in the chance of a report and the increase in the fix rate overall predicted by the model. To think about the difference between these two numbers, a large relative increase on something that has a low base rate of repair might lead to fewer new fixes than a relatively small relative increase on a category with a high fix rate. These generally track in each in size, with minor differences in order depending if sorted by relative or absolute change.

The categories with the largest absolute increases as a result of a photo are Highways Enquiries (11.5 percentage points increase) and reports in Open Spaces/Parks (9.7 percentage points). The labelled categories with  the highest negative scores are rights of way (minus 4 percentage points) and pavement/footway defects (minus 2.4 percentage points). Reflecting the fact that the overall picture across all reports is that photos had a positive effect, there are fewer categories with a negative effect and these effects tend to be smaller.

While there is a clear reason to expect positive effects (photos are helpful in identifying and locating problems), the cause of these negative categories is less clear. One possible explanation for this is if a photo is meant to help triage and prioritise problems, this is exactly what is happening but the report is being given a low priority. From the authority’s point of view, being able to evaluate an issue as low-priority from a photo is a good thing.

In other cases, photos may be especially unhelpful at identifying issues. The category with the highest relative decrease is Street Lights (minus 40.5% percentage chance of a fix). One possible explanation for this is that street light problems are most noticeable when it is dark, so may be especially hard to take useful pictures of.


Attaching a photo of a report in most cases improves the odds of getting a fix, but this varies by the kind of report. There is a set of categories where a photo has positive effects of varying strengths, and a smaller group where there is a small negative effect. In terms of impacts on the service, when photos can convey useful information that help lead to a resolution, users should be encouraged to attach them. Where photos are  less helpful, other interface or prompt suggestions for more relevant information or asset selection may help lead to more repairs.

Category Relative change Absolute change
Highways Enquiries 69.8% [56.1%, 84.6%] 11.2% [5.4%, 14.6%]
Open Spaces/Parks 49.6% [34.8%, 66.1%] 9.7% [7.0%, 12.4%]
Incivilities 64.1% [50.5%, 79.0%] 8.4% [5.8%, 11.3%]
Parking 41.1% [20.8%, 64.7%] 4.9% [1.9%, 8.9%]
Road Safety 20.7% [12.5%, 29.4%] 4.6% [2.9%, 6.3%]
Dog Fouling 34.6% [11.1%, 63.1%] 3.3% [1.0%, 6.5%]
Overgrown/Fallen Veg/Trees 20.0% [13.3%, 27.0%] 3.2% [2.1%, 4.4%]
Rubbish 28.0% [24.8%, 31.2%] 2.8% [2.4%, 3.3%]
Abandoned Vehicles 25.6% [16.3%, 35.7%] 1.2% [0.6%, 2.0%]
Road Surface Defects -5.0% [-7.6%, -2.3%] -1.2% [-1.9%, -0.6%]
Street Lights -40.5% [-43.7%, -37.1%] -1.8% [-2.4%, -1.3%]
Pavement /Footway Defects -10.3% [-15.2%, -5.0%] -2.4% [-3.8%, -1.1%]
Right of Way -15.3% [-22.0%, -8.0%] -4.0% [-6.0%, -2.0%]

Table 1 – Relative and absolute change by category (sorted by absolute)

Data source and analysis scripts available on GitHub

Image: Hendrik Morkel