NKM fig 3

Housing Licensing: 1) Needs assessment

[Update 15 April 2014: Scroll down to comments for a more detailed explanation of why the statistical report is flawed.]

The basis for Enfield Council’s private rented sector licensing proposal is quantitative research conducted by Mayhew Harper Associates (MHA/NKM), which has attempted to establish a link between the private rented sector and anti-social behaviour. This has been combined with qualitative research (focus groups etc.) run by  Opinion Research Services (ORS) to identify the need.

The executive summary of the ORS report (full report not available) states at paragraph 1.12 that:

In the residents’ forums opinion was divided over whether there is a link between the PRS and ASB or whether problems were simply the result of an overall increase in population.

[PRS: private rented sector; ASB: anti-social behaviour]

The MHA/NKM report is based on modelled data which has attempted to predict which houses are in the private sector (either single family dwellings or houses in multiple occupation). No validation results are presented (assuming, I hope, that some was done).

The report does not include a comparison of anti-social behaviour for different tenure types (private rented, social rented, owner-occupier). The one straight line chart presented  covers only modelled single family dwellings and does not even have a statistical value displayed to show the strength (or weakness) of the correlation (the R2 coefficient).

NKM fig 6

The Council also used maps produced from the modelled data as evidence of a correlation. They overlaid modelled PRS home locations on ASB rate maps:

NKM fig 3


What they didn’t do was any kind of comparison with the other social determinants of crime. I looked at Enfield’s own public health website and obtained this map of the index of multiple deprivation by ward:


This replicates their map quite well (partly because one of the deprivation factors is crime rate). This shows that there is probably more than one underlying driver of anti-social behaviour. There is also a stark west-east divide, with the dividing line being the route of the A10 road.

Even if MHA/NKM could demonstrate a simple correlation with just one factor (tenure type), this is not proof of causation. The report itself acknowledges that:

Note that because we have not uniquely ascribed ASB to individual households (for reasons previously given) this does not necessarily demonstrate cause and effect.

I question the basis for their conclusion that:

Enfield Council’s hypothesis that privately rented properties are associated with high levels of ASB also appears to have reasonable justification. In the limited instances where addresses were linkable, higher than average percentages of ASB were attributable to higher risk privately rented households.

In an earlier section they find that “statistically the correlation is weak”, so their conclusion seems baseless. They were only able to link ASB incidents to addresses in about 5% of cases, so the final sentence should be disregarded. Their conclusions are not supported by their own work and yet they have managed to make this bold claim, which Enfield Council have used to underpin their policy.

6 thoughts on “Housing Licensing: 1) Needs assessment”

  1. wow most of that is too clever for me!!!!!!! but just 5 % of asb incidents matched to addresses thats small numbers to justify a licensing scheme

  2. Hi Graham

    Yes, you are right on both issues.

    I kept my comments about the graph short in my original post but there are many things wrong with it:

    On the vertical (y) axis the graph includes data about one tenure type (single family households), divided by total number of properties. This is not good because there is no comparison with other tenure types (HMO, social housing, owner-occupier, live-in landlord). Also, dividing by number of properties is misleading because properties may include more than one household (a household is defined as members of a family or those living as a couple). They would have been better off dividing dwelling tenure types by the number of dwellings (maybe this is what they did, but it isn’t clear – the paragraphs below say that they divided properties by properties – either way, the lack of clarity around what they have done is sloppy).

    Turning to the horizontal (x) axis, this is also problematic. It shows incidents of ASB per property. Does this make sense? Properties are inanimate. They don’t cause anti-social behaviour. A far better measure would have been incidents per head of population.

    Now, addressing your points:

    The percentages on the vertical axis are quite small, so a tiny error in their modelling could have made a big difference here. On the horizontal axis, there is no indication of what the range of households is. Tiny movements in recorded ASB could have made an enormous difference. I’m thinking of examples where grid cells overlap with the reservoirs, for example. The academic literature has methods for dealing with this kind of issue but that approach hasn’t been used here.

    If a straight line was a good model for this, the data would have been much closer to the line. If the outliers had been included (the little red outlined boxes), the straight line would have been a really bad fit.

    Just eyeballing the data points, we can also see that if the straight line had been extended, it wouldn’t go through the origin (0,0). This chart would therefore predict that if single family households were about 0.3% there would be no antisocial behaviour in the borough (or in these tenture types, again it’s unclear what has been done here). Is that good or bad? We don’t know. In theory it’s possible to have no households but for ASB to happen in that area.

    Another observation is that as we move from left to right, the data spreads out. In grid cells where 20 incidents of ASB are observed per property, there are between 1.0% and 4.4% of predicted single family households. This suggests a pretty poor fit but we await to see whether Enfield Council will publish the data.

    What’s the significance of data spreading out (“heteroscedasticity” in stats speak)? If we were to compare this hypothesis with those for other tenure types and took the spread of the data into account, it might overlap with the variation from the other tenure types. That would mean rejecting the hypothesis that there is a link between private rented sector properties and ASB… food for thought…

    The above would all have been huge improvements if they had been able to match ASB to actual property addresses, with known tenure types. They were only able to match 5% of incidents to addresses, so the whole exercise is flawed unless they did some validation. The report doesn’t say that they did any checking.

    Hopefully that has answered your points, and shows why I didn’t go into detail on the original post!

    Loose Minute

  3. I’m not a mathematician, does it mean that if the data is strong all the points would be on the line or very close to it ?……… and the percentages on the x axis seem to be very low values?

Leave a Reply

Your email address will not be published. Required fields are marked *