## Introduction

This article will try to answer the question if a regions Brexit vote determines adherence to lock-down. In very simple terms this means if a region has voted for Brexit with a higher percentage does show more mobility.

Many scientific studies have looked at Brexit. Reasons listed are:

Sovereignty

Immigration

Age of voters

Education level

Order vs openness

The “left behind”

Britons felt less integrated into the EU

Identity and Change

English national identity

Anti-establishment populism

Since all this research classifies the “**classic Brexit voter**” as an old, low educated, xenophobic, anti-establishment, English nationalist populist, it would be interesting if these people show a different behaviour to younger, higher educated, inclusive, establishment friendly non English nationalist reasonable voters. Please see the ecological fallacy and Robinson’s paradox.

So because of the ecological fallacy and Robinson’s paradox the only thing we will be looking here is if the extent of the Leave vote in a region determines (**the number not the voters**) if these regions behaved differently on the target date. Is there a Brexit Vote lock-down connection?

Taking Google’s mobility data from the 1st of May 2020 and adding the Brexit vote brought interesting results. More details about the data is explained here.

Basically we want to determine in the categories: Retail and Recreation, Grocery and Pharmacy, Parks, Workplaces, Transit stations and Residential Areas if the Brexit voting regions behave differently.

These categories have been decided by Google. They classify their categories as such.

**Retail & Recreation** are places like restaurants, cafes, shopping centres, theme parks, museums, libraries, and movie theatres.

**Grocery & Pharmacy** are places like grocery markets, food warehouses, farmers markets, speciality food shops, drug stores,

and pharmacies.

**Parks** are defined as national parks, public beaches, marinas, dog parks, plazas, and public gardens.

**Transit** **stations** are public transport hubs such as subway (the Underground or Metro) , bus/coach, and train stations.

**Workplaces** are places were people work.

**Residential** are places were people live or reside.

Data from the last date that Google recorded (at the time of writing this article this was the 26th of April 2020) was taken or as Google does it from the last date that data was available. Google defined 86 English areas, 11 Northern Ireland Regions, 32 Scottish Councils as Regions and 22 regions in Wales.

The Brexit vote data for each region was taken from Wikipedia and the BBC. The Brexit vote will be represented as a fraction. So 50% would be 0.5.

## Results

### Retail, Recreational and Leave (Brexit) Vote

The histograms of the Retail values for the 12/04/2020 and the 26/04/2020 look reasonably Gaussian.

`Linear Regression 12/04/2020`

lm(formula = Retail ~ Leave, data = subset(mobdf, UKSUB == "England" |

UKSUB == "Wales"))

Residuals:

Min 1Q Median 3Q Max

-11.7221 -2.1508 -0.4982 1.5770 12.8634

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) -91.995 2.751 -33.441 < **2e-16 *****

Leave 21.180 4.954 4.275 **4.22e-05 *****

---

Signif. codes: **0 ‘***’** 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 3.752 on 105 degrees of freedom

Multiple R-squared: 0.1483, Adjusted R-squared: **0.1402**

F-statistic: 18.28 on 1 and 105 DF, p-value: 4.216e-05

`Linear Regression 26/04/2020`

lm(formula = Retail ~ Leave, data = subset(mobdf, UKSUB == "England" |

UKSUB == "Wales"))

Residuals:

Min 1Q Median 3Q Max

-14.0341 -1.6572 -0.3284 1.4229 16.7454

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) -88.822 2.990 -29.702 < 2e-16 ***

Leave 19.478 5.382 3.619 **0.000455 *****

---

Signif. codes: **0 ‘***’** 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 4.083 on 106 degrees of freedom

Multiple R-squared: 0.11, Adjusted R-squared: **0.1016**

F-statistic: 13.1 on 1 and 106 DF, p-value: 0.0004545

Retail Mobility seem to vary with the Brexit vote fraction in England and Wales on the 12/04/2020 and in England on the 26/04/2020. The variation explained seems to be fairly low.

### Grocery, Pharmacies and Leave (Brexit) Vote

`Linear Regression 12/04/2020`

Call:

lm(formula = Grocery ~ Leave, data = subset(mobdf, UKSUB == "England"))

Residuals:

Min 1Q Median 3Q Max

-10.1569 -2.3605 -0.1465 2.1521 9.1907

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) -40.586 3.013 -13.470 < 2e-16 ***

Leave 14.858 5.379 2.762 **0.00706 ****

---

Signif. codes: **0 ‘***’** 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 3.812 on 83 degrees of freedom

Multiple R-squared: 0.0842, Adjusted R-squared: **0.07317**

F-statistic: 7.631 on 1 and 83 DF, p-value: 0.007063

`Regression 26/04/2020`

Call:

lm(formula = Grocery ~ Leave, data = subset(mobdf, UKSUB == "England"))

Residuals:

Min 1Q Median 3Q Max

-7.8556 -2.4205 -0.2156 1.9289 13.6345

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) -46.414 2.809 -16.521 < 2e-16 ***

Leave 17.309 5.012 3.454 **0.000869 *****

---

Signif. codes: **0 ‘***’** 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 3.557 on 84 degrees of freedom

Multiple R-squared: 0.1243, Adjusted R-squared: **0.1139**

F-statistic: 11.93 on 1 and 84 DF, p-value: 0.0008686

Grocery Mobility seem to vary with the Brexit vote fraction in England on the 12/04/2020 and on the 26/04/2020. The variation explained seems to be fairly low. Looking at the analytics plots this model seems to fit a lot better with less long tails. An exception is Luton. The variation explained with the Leave vote is between 7 and 11%.

### Parks and Leave (Brexit) Vote

From the pictures it seems that there is not enough detail in the differences between Regions and the influence of the Leave vote.

### Transit and Leave (Brexit) Vote

`Linear Regression 12/04/2020`

Call:

lm(formula = Transit ~ Leave, data = subset(mobdf, UKSUB == "England"))

Residuals:

Min 1Q Median 3Q Max

-23.2865 -4.4497 -0.5246 3.8378 19.2781

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) -100.996 6.216 -16.248 < 2e-16 ***

Leave 62.448 11.084 5.634 **2.41e-07 *****

---

Signif. codes: **0 ‘***’** 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 7.837 on 82 degrees of freedom

(1 observation deleted due to missingness)

Multiple R-squared: 0.2791, Adjusted R-squared: **0.2703**

F-statistic: 31.74 on 1 and 82 DF, p-value: 2.41e-07

`Linear Regression 26/04/2020`

Call:

lm(formula = Transit ~ Leave, data = subset(mobdf, UKSUB == "England"))

Residuals:

Min 1Q Median 3Q Max

-32.536 -6.949 -0.928 6.207 27.668

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) -98.974 8.451 -11.711 < 2e-16 ***

Leave 71.699 15.060 4.761 **8.07e-06 *****

---

Signif. codes: **0 ‘***’** 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 10.66 on 83 degrees of freedom

(1 observation deleted due to missingness)

Multiple R-squared: 0.2145, Adjusted R-squared: **0.205**

F-statistic: 22.67 on 1 and 83 DF, p-value: 8.069e-06

Transit Mobility seem to vary with the Brexit vote fraction in England on the 12/04/2020 and on the 26/04/2020. Looking at the analytics plots this model seems to fit a lot better with less long tails. An exception is Luton. The variation explained with the Leave vote is between 20% and 25%, which means that up to 25% of the Transit places variation can be explained by the Leave vote fraction.

### Workplaces and Leave (Brexit) Vote

`Linear Regression 12/04/2020`

Call:

lm(formula = Workplaces ~ Leave, data = subset(mobdf, UKSUB ==

"England"))

Residuals:

Min 1Q Median 3Q Max

-4.227 -1.363 0.139 1.417 7.491

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) -72.262 1.743 -41.451 < 2e-16 ***

Leave 29.181 3.112 9.377 **1.15e-14 *****

---

Signif. codes: **0 ‘***’** 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 2.206 on 83 degrees of freedom

Multiple R-squared: 0.5144, Adjusted R-squared: **0.5085**

F-statistic: 87.92 on 1 and 83 DF, p-value: 1.152e-14

`Linear Regression 26/04/2020`

Call:

lm(formula = Workplaces ~ Leave, data = subset(mobdf, UKSUB ==

"England"))

Residuals:

Min 1Q Median 3Q Max

-8.3573 -1.8843 0.0073 1.5714 9.0859

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) -68.565 2.489 -27.549 < 2e-16 ***

Leave 40.196 4.440 9.053 **4.63e-14 *****

---

Signif. codes: **0 ‘***’** 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 3.151 on 84 degrees of freedom

Multiple R-squared: 0.4939, Adjusted R-squared: **0.4878**

F-statistic: 81.96 on 1 and 84 DF, p-value: 4.635e-14

Workplace Mobility seem to vary with the Brexit vote fraction in England on the 12/04/2020 and on the 26/04/2020. Looking at the analytics plots this model seems to fit OKish. An exception is Luton. The variation explained with the Leave vote is between 48 % and 50%, which means that up to 50% of the Workplace variation places can be explained by the Leave vote fraction.

### Residential and Leave (Brexit) Vote

From the pictures it seems that there is a nonlinear relationship between the differences between Regions Residential Areas and the influence of the Leave vote.

## Summary

As this as an ongoing epidemic it was interesting if the two above glm snapshots on the 12/04/2020 and on the 26/04/2020 represented a long standing phenomenon. It turns out they are.

Running above model over time reveals that this different behaviour is long running and started in March.

## Conclusion

As with all statistics, we deal with uncertainty, so no result is ever perfect. Some of the residual plots in the above models are not ideal. Nevertheless Minitab (a statistics software) made an analysis.

“If you have nonnormal residuals, can you trust the results of the regression analysis?” Link

The answer was:

The study found that a sample size of at least 15 was important for both simple and multiple regression. If you meet this guideline, the test results are usually reliable for any of the non-normal distributions.

As such the result should be pretty robust. We have 6 mobility classifications by Google. In 4 of these we find some form of connection between the amount of the Leave vote and how well that region in England locked down. That is 66.66% of the results. **And this result still remained two weeks later**. As one knows correlation does not imply causation, so we can not directly assume that the Brexit vote is the causation of the adherence to lock-down. But as we know the Brexit vote has been linked to the Age of voters, their

Education level, The “left behind”and English national identity.

So it is entirely possible that the difference between regions is reflected in this. But older workers would most likely, due to retirement, be less likely to go to work, so this is astonishing.

Transit places are also linked to work places as they are needed to go to work. Why the “older left behind” as the Guardian describes them would go more to work is essentially unclear. Either the conclusion is wrong that it was the older left behind or the “uneducated” are more likely to be key-workers, or part of the gig economy and felt forced to go to work. Some of course might have also resisted the lock-down out of principle.** But we do not know and can not conclude that is the Pro Brexit voters driving that trend.** It could well be that regions with more older people need more carers and so it could be that these groups drive this trend.

It is also possible that differences are driven through the country city dichotomy of the Brexit vote as country users would need to cover bigger distances to get food and messages.

If we assume that Brexit voting regions are more right wing, due to “English nationalism“, one could conclude that right wing places are more anxious about the economy and would as such maybe comply less.

There are right wing groups, at least in the United States, that would like the lock-down to be lifted.

In the moment it is unclear why the connection between the Brexit voting regions and some mobility aspects could exist.