ITF0001

 

Written evidence submitted by Chris Hanretty

 

  1. I am writing to provide evidence concerning the selection of towns for the Towns Fund.
  2. I am Professor of Politics at Royal Holloway, University of London. I have experience in modelling election outcomes and combining data from different administrative geographies.
  3. The essence of my submission to the committee is that the process by which towns were invited to bid for money from the Towns Fund was driven by party-political electoral advantage.
  4. Specifically, considering the 541 towns judged eligible for the scheme as a whole:
    1. Towns located in whole or in part in Westminster constituencies held by the Conservative party were ten percentage points more likely to be selected than all other eligible towns;
    2. Towns located in marginal Conservative seats (majorities of greater than five percent but less than ten percent) were 37 percentage points more likely to be selected than towns located in safe seats held by other parties. Towns located in ultra-marginal Conservative seats (majorities of less than five percent) were 21 percentage points more likely to be selected than towns located in safe seats held by other parties.
    3. The success rate for “low priority” towns in Conservative seats was higher than the success rate for “medium priority” towns not held by the Conservative party.
    4. Party-political differences in the rate at which towns were selected persist after controlling for the ranking of towns produced by servants; the component scores making up that ranking; the population of each town; the region of each town; and whether the town was part of a local enterprise partnership which was well- or poorly-represented within the region.
  5. I make these claims on the basis of the data contained in the National Audit Office report “Review of the Town Deals selection process”. This report contains information on component scores for each town, overall rankings, and whether each town was selected. The report also contains ONS geography codes for each town. I have used these codes, and boundary files for built-up areas and Westminster constituencies, to work out which towns are contained in which Westminster constituencies.
  6. I analyse this data in two ways. First, I produce charts of success rates by different characteristics of the towns selected. Second, I use statistical models to predict the probability of selection given different town characteristics.
  7. Figure 1 shows the success rate of towns according to different political factors, and bears out the claims given in point (4) above. I use “Conservative-held town” to refer to a town which was located, in whole or in part, in a Westminster constituency won by the Conservative party in the general election of 2017. When I refer to Conservative majorities, I refer either to the Conservative majority in the relevant seat (for towns wholly contained within a single constituency) or to the average Conservative majority in the relevant seats (for towns which span constituency boundaries). The top panel includes data on all 514 towns. The middle panel excludes three towns located in part in the Speaker’s seat. The bottom panel excludes “high priority” towns, where ministers accepted civil servants’ recommendations.
  8. Table 2 shows the results of five linear regression models of town selection. Each model (columns 1 – 5) includes a different number of control variables (which appear down the rows). Each cell value gives the effect associated with the variable in that row, and the confidence interval associated with that effect. Thus, the entry in the second row of the first column means “the effect of a town being in a Conservative-held seat was to increase the probability of selection by 9.9 percentage points, but this effect could well be anywhere between 2.9 percentage points and 16.9 percentage points”.
  9. The results of the models show that there was a statistically significant effect of being in a Conservative held town, and that this effect persists when controlling for the other predictors discussed in point 4(d) above. The effect ranges between nine and sixteen percentage points.
  10. Further details regarding these models and details regarding additional models not discussed here are supplied in an additional paper prepared with a view to academic publication.
  11. Although reasonable people can disagree about the appropriate regression modelling strategies, it is my professional judgement that no reasonable analyst of the NAO data could fail to conclude that the selection process was affected by party-political considerations.

 


Table 1: Regression model of town selection

 

 

 

(A1)

(A2)

(A3)

(A4)

(A5)

Intercept

0.120

0.248

-0.193

-0.196

-0.211

 

[0.062, 0.178]

[0.187, 0.310]

[-0.314, -0.071]

[-0.317, -0.074]

[-0.365, -0.057]

In Conservative seat

0.099

0.114

0.087

0.090

0.161

 

[0.029, 0.169]

[0.049, 0.180]

[0.024, 0.149]

[0.028, 0.153]

[0.079, 0.242]

Adjusted rank

 

0.006

0.004

0.004

0.004

 

 

[0.005, 0.007]

[0.003, 0.006]

[0.003, 0.006]

[0.002, 0.005]

log(Pop’n)

 

 

0.135

0.138

0.142

 

 

 

[0.102, 0.168]

[0.105, 0.171]

[0.107, 0.176]

LEP concentration

 

 

 

-0.077

-0.074

 

 

 

 

[-0.170, 0.015]

[-0.166, 0.019]

East of England

 

 

 

 

-0.121

 

 

 

 

 

[-0.253, 0.010]

North East

 

 

 

 

-0.029

 

 

 

 

 

[-0.155, 0.097]

North West

 

 

 

 

-0.001

 

 

 

 

 

[-0.110, 0.108]

South East

 

 

 

 

-0.168

 

 

 

 

 

[-0.278, -0.058]

South West

 

 

 

 

-0.088

 

 

 

 

 

[-0.202, 0.026]

W Midlands

 

 

 

 

-0.090

 

 

 

 

 

[-0.210, 0.029]

Yorks and Humber

 

 

 

 

0.052

 

 

 

 

 

[-0.067, 0.172]

Num.Obs.

541

541

541

541

541

R2

0.014

0.135

0.229

0.233

0.257

R2 Adj.

0.012

0.132

0.225

0.227

0.242

AIC

513.9

445.3

384.7

384.0

380.6

BIC

526.8

462.4

406.1

409.7

436.4

Log.Lik.

-253.953

-218.625

-187.329

-185.982

-177.301


Figure 1:

 

September 2020