Impact of occupations on UK rentals

Assessing if (and how) occupations affect local rental market
random effects
hierarchical models
glm
binomial
Author

Prateek

Published

November 30, 2025

Data

Cross tabulated MSOA level data on Tenure by Occupation, from ONS 2021 Census can be found from Nomisweb API (“RM140: Tenure by Occupation - Household Reference Persons” 2023)
The unit of measurement (i.e. a single count) refers to a Household Representative persion (referred to as HRP henceforth), thus associating a unique occupation to a given household. This data will differ from total count of people in full time occupation from any other dataset (e.g. data on occupation by age, as used in this study), because there may be multiple people in full time occupation in a household.

2021 super output area - middle layer Total 1. Managers, directors and senior officials 2. Professional occupations 3. Associate professional and technical occupations 4. Administrative and secretarial occupations 5. Skilled trades occupations 6. Caring, leisure and other service occupations 7. Sales and customer service occupations 8. Process, plant and machine operatives 9. Elementary occupations Tenure of household
E02006012 : Shropshire 033 689 156 171 105 31 103 26 17 48 32 Owns with a mortgage or loan or shared ownership
E02005850 : Broxtowe 001 253 25 32 31 22 34 31 16 32 30 Private rented or lives rent free
E02006167 : Newcastle-under-Lyme 010 490 39 92 48 38 53 57 43 51 69 Private rented or lives rent free
E02000054 : Barnet 031 91 8 19 7 16 4 10 6 11 10 Social rented
E02004237 : Bournemouth, Christchurch and Poole 014 451 81 85 66 44 45 37 32 36 25 Owns outright
W02000352 : Newport 006 317 12 23 27 26 28 70 29 43 59 Social rented

Dataset

Additional information

Local authority for each MSOA in original dataset is added, to account for rental patterns that may be present at local area levels. Being in the same Local authority may affect neighbouring MSOAs introducing correlations in their data.

It can be possible that age has a confounding effect on Rentership via Occupation. Majority of Professional and Managers may likely be older than associates and this may mean that they were able to purchase a house earlier in their careers/life and now represent a smaller proportion of rental cohort.
Adjusting for age gives an apples-to-apples comparison of rentership preference among people pursuing different occupations but are in boradly same phases of their life.
The adjustment is made by including the proportion of full-time employed HRPs in an MSOA engaged in a particular profession falling in one of the following age bands:

  • aged 15 years and under
  • aged 16 to 24 years
  • aged 25 to 34 years
  • aged 35 to 49 years
  • aged 50 to 64 years
  • aged 65 years and over

Note:
This adjustment is inexact since the data is available for all people in full time occupation rather than just HRP (which is the unit of measurement in tenure data). This means that likely an HRP would be an older person, the counts in age dataset include other members of the household which are possibly younger, creating a slight disconnect between the 2 datasets.
It is neverthless instructive to use this information as the downsides due to excluding this information outweigh downsided due to (incorrectly) including it.

Since, the interest lies in proportion of rentership for each occupation, per occupation totals for each MSOA are added to the dataset.

Filter

Current data only includes measurements from household reference persons (HRP) in full time employment a week before census.
It would be sensible to drop data if any occupation icluded positive entries in aged 15 years and under column.

Note:

msoa tenure occupation counts_of_hrp occupation_total aged 15 years and under aged 16 to 24 years aged 25 to 34 years aged 35 to 49 years aged 50 to 64 years aged 65 years and over all_ages_total region lad_name
e02005269 private rented or lives rent free 4. administrative and secretarial occupations 115 229 0 0.0166945 0.0287980 0.0219115 0.0169032 0.0012521 4792 e12000002 wyre
e02006338 private rented or lives rent free 1. managers, directors and senior officials 46 344 0 0.0026898 0.0194262 0.0833831 0.0517035 0.0050807 3346 e12000008 epsom and ewell
e02003540 private rented or lives rent free 6. caring, leisure and other service occupations 61 215 0 0.0125856 0.0242879 0.0342239 0.0275999 0.0041952 4529 e12000008 portsmouth
e02006410 private rented or lives rent free 6. caring, leisure and other service occupations 38 169 0 0.0147865 0.0202719 0.0379203 0.0264727 0.0035774 4193 e12000008 spelthorne
e02002510 private rented or lives rent free 8. process, plant and machine operatives 34 212 0 0.0050837 0.0161483 0.0281100 0.0325957 0.0044856 3344 e12000003 north yorkshire
e02004298 private rented or lives rent free 1. managers, directors and senior officials 33 416 0 0.0024234 0.0154219 0.0623485 0.0552985 0.0090328 4539 e12000001 county durham

Subsample

Since, the dataset is very well compiled but is large, we can work with reduced dataset and infer large scale effects from it. This reduction will allow speeding up the analysis and experimentation.
The subsampling process does the following:

  • Randomly sample 10 local authorities from 8 UK wide regions (e.g. West midlands, East of England etc.) excluding London region (since market behaviour may not generalise to other regions and create issues when data is mixed with others, due to very high counts)
  • Within each of the sampled local authorities sample 50 MSOAs at random
  • Get all the data in the sampled MSOAs

The above procedure is deliberate in that it results in balanced data among Local authorities to avoid issues with inference.
It goes without saying that this step means that Local Authorities and MSOAs need to be treated as random effects down the line (with appropriate heirarchy).
The procedure ensures sufficient counts of Local Authorities and MSOAs to be able to reliably capture these random effects.

rows distinct_tenure distinct_occupation max_counts min_counts median_counts distinct_lad distinct_msoa
18252 1 9 969 1 40 80 2028

EDA

At the level of MSOA the proportion of rentership can appear quite chaotic and noisy.

The variability in renting due to occupation is visible when looked at in aggregate. This also seems to relate to our intuition where highly skilled and professionals are driving less of the rental market since they may be better placed to afford to buy.

Warning: Removed 558 rows containing non-finite outside the scale range
(`stat_boxplot()`).

The observation holds and patterns become more clear when the data is aggregated at Local authority level. This means that a heirarchical consideration of geographies could be a sensible inclusion in the model.

`stat_bin()` using `bins = 30`. Pick better value `binwidth`.

There is a slightly long tail of working population per msoa.

Looking at the msoa where the greatest proportion of working population is renting, a few key dense local authority areas are highlighted

lad_name msoa aggregate_proportion_renting total_working_population
bradford e02006948 0.728 1660
leeds e02002373 0.719 2022
leeds e02006875 0.712 2797
leeds e02002392 0.711 1508
leicester e02002849 0.694 3566
east riding of yorkshire e02002666 0.677 2259
southampton e02003565 0.650 1931
leicester e02002842 0.641 2504
southampton e02003571 0.627 2236
peterborough e02003250 0.623 3272

Modelling

Baseline Model: Fixed effects with data aggregated at LAD level

We refer back to data aggregated at LAD level previously. The variability at each occupation level is quite modest and for the sake of simplification LAD variable is ignored in the baseline. It is expected that this may not be systematically related to other variables like age and occupation.
Since the data is further aggregated, entire dataset can be used disregarding the need to subsample. Further, Elementary occupations are treated as a reference category and effects of other categories are in contrast to this.
Occupation variable appears to be significantly related to probability of renting when controlled for age. 

Analysis of Deviance Table

Model 1: cbind(renting_total, occupation_total) ~ `aged 25 to 34 years` + 
    `aged 35 to 49 years` + `aged 50 to 64 years` + `aged 65 years and over`
Model 2: cbind(renting_total, occupation_total) ~ occupation + `aged 25 to 34 years` + 
    `aged 35 to 49 years` + `aged 50 to 64 years` + `aged 65 years and over`
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      2101     114755                          
2      2093      57716  8    57039 < 2.2e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

The diagnostics do not look too bad and we can retain the model for comparison with more detailed models.

Additionally, the coefficients make intuitive sense, each occupation category is likely to have less association with renting compared to elementary occupations.
Concretely, people in professional occupations are ~50% less likely (exp(-0.70)-1) to rent compared to people in elementary occupations.

Heirarchical Binomial model at MSOA level with

Alternate formulation with Poission rentership and Housholds as exposure

TODO

References

“Annex Table 1.5: Employment Status by Tenure,2021-22.” 2023. Department for Levelling Up, Housing; Communities. https://assets.publishing.service.gov.uk/media/64ad65e48bc29f000d2cca7f/EHS_21-22_PRS_Ch_1_Annex_Tables.ods.
“English Housing Survey 2021 to 2022: Private Rented Sector.” 2023, July. https://www.gov.uk/government/statistics/english-housing-survey-2021-to-2022-private-rented-sector/english-housing-survey-2021-to-2022-private-rented-sector.
“RM140: Tenure by Occupation - Household Reference Persons.” 2023. Office for National Statistics. https://www.ons.gov.uk/datasets/RM140/editions/2021/versions/2.