search

Leeds-MRG / Minos

SIPHER Microsimulation for estimating the effect on Income policy on mental health.
MIT License
4 stars 3 forks source link

Updating Transition Models #103

Closed ld-archer closed 1 year ago

ld-archer commented 1 year ago

Opening this issue to track the decisions and changes made in improving transition models.

hh_income

Following a meeting with SIPHER members (9/12/21), we had a couple of ways to improve the hh_income model. Following these suggestions, we have made some changes to the model that are pragmatic but could be improved upon in the future.

  1. Lag hh_income The lag of hh_income has been added to the model. There is debate as to whether we should use the lag of something to predict its next state, but for now it is fine. This could be replaced however with something like the Random Effects (2.2) or Fixed Effects (2.3) from this document.

  2. Static job_sec and job_sector These 2 variables are held static, and not transitioned over time. This decision was made to reduce complexity, as it means we do not need to predict the next state of each of these things before predicting next state of hh_income. If we could predict these variables effectively however, we would most likely get a more robust/'accurate' predictive model (accurate is probably the wrong word here).

  3. Labour state Labour state is no longer transitioned, and therefore is not included in the hh_income model. Again this would most likely improve the hh_income model, but also adds complexity. If necessary in the future, we can either transition this variable separately from job_sec or we could combine the two, adding additional levels to the 8 level NSSEC (retired, student, unemployed etc.)

  4. Household information At present, no information on other members of the household is included in the prediction of hh_income. This could improve things, but again adds a lot of complexity (more than the previous 3 points). We would have to figure out how to transition households into the future, as well as predict all the things we want to include in these models. Lets avoid that for now.

alcohol_spending

This model has been removed, as the impact of alcohol spending (or consumption for that matter) on mental health wasn't clear enough.

housing_quality

The previous version of the composite had three levels:

  1. Access to all identified components
  2. Access to some
  3. No access

This was a problem as the composite was heavily skewed - almost nobody had no access to the components, and movement between the levels was very unbalanced (i.e. moving from 3 to 2 had a much bigger impact than moving from 2 to 1). Instead we have identified a core set of components that are important for housing quality, and a 'bonus set' for want of a better term. The core set are:

Which leaves for the bonus set:

The new composite will have 5 levels depending on access to core/bonus variables:

  1. Missing 1+ core components
  2. All core and some bonus components
  3. All core and all bonus components

Will test the distribution of this composite after generating and post results here.

ld-archer commented 1 year ago

Housing Quality (cont.)

This is the distribution of the new composite. Much more balanced, where the most common group is level 2 (all core some bonus), and both other groups have a decent chunk of people. image

Now worth testing a transition model for the relationship between hh_income and the new composite:

Minimal model

> summary(housing)
formula: housing_quality ~ scale(hh_income)
data:    data

 link  threshold nobs     logLik       AIC         niter max.grad cond.H 
 logit flexible  44407783 -38980356.07 77960718.14 6(0)  1.75e-08 3.2e+00

Coefficients:
                  Estimate Std. Error z value Pr(>|z|)    
scale(hh_income) 0.5195491  0.0004357    1193   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Threshold coefficients:
      Estimate Std. Error z value
1|2 -2.4353675  0.0005439   -4477
2|3  0.6576626  0.0003217    2044
(255 observations deleted due to missingness)

Full Model

> summary(housing)
formula: housing_quality ~ scale(age) + factor(sex) + scale(SF_12) + relevel(factor(ethnicity), ref = "WBI") + scale(hh_income)
data:    data

 link  threshold nobs     logLik       AIC         niter max.grad cond.H 
 logit flexible  44407783 -38385771.83 76771577.67 6(0)  9.21e-08 1.8e+03

Coefficients:
                                             Estimate Std. Error z value Pr(>|z|)    
scale(age)                                 -0.2072656  0.0003146 -658.81   <2e-16 ***
factor(sex)Male                             0.0399841  0.0006028   66.33   <2e-16 ***
scale(SF_12)                                0.1902093  0.0004009  474.44   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")BAN -1.3238875  0.0040016 -330.84   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")BLA -1.1783021  0.0025514 -461.82   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")BLC -1.0926503  0.0036256 -301.37   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")CHI -0.9654861  0.0047377 -203.79   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")IND -0.4802789  0.0019294 -248.93   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")MIX -0.2987210  0.0023905 -124.96   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")OAS -1.1041091  0.0028327 -389.78   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")OBL -1.1764196  0.0115136 -102.18   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")OTH -0.2691478  0.0048285  -55.74   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")PAK -0.8678519  0.0024983 -347.38   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")WHO -0.1683221  0.0012785 -131.65   <2e-16 ***
scale(hh_income)                            0.4999786  0.0004391 1138.72   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Threshold coefficients:
      Estimate Std. Error z value
1|2 -2.5228946  0.0006476   -3896
2|3  0.6460611  0.0004562    1416
(255 observations deleted due to missingness)
ld-archer commented 1 year ago

Loneliness

Smoking and drinking intensity have been removed from this model.

Marital status and household composition (see below) have been added as predictors.

We were also asked to justify ethnicity as a predictor of loneliness, which I believe this paper does (Franssen et al (2020).

The association between ethnicity and loneliness was stronger among young and early middle-aged adults, compared to late middle-aged adults.

This wasn't the only paper I found that mentioned a link, we can get more if necessary.

Household Composition

A representation of household composition has been added to use as a predictor of loneliness. To generate this variable, we reduced and simplified a household composition variable from Understanding Society (hhtype_dv). The original variable had 18 levels, we reduced this to 4 levels:

  1. Single adult no kids
  2. Single adult 1+ kids
  3. Multiple adults no kids
  4. Multiple adults 1+ kids

Counts by group:

3    15119
4     9589
1     4959
2      785

hh_composition_counts

Marital Status

The 9 level marstat variable from Understanding Society has 9 levels, but some covering less than 1% of the sample. We have recoded the variable into 4 levels:

  1. Single never partnered
  2. Partnered a. Married b. Civil Partner c. Living as couple
  3. Separated a. Separated legally married b. Divorced c. Separated from civil partner d. A former civil partner
  4. Widowed a. Widowed b. Surviving civil partner
Partnered    19597
Single        6420
Separated     2497
Widowed       1859
-9              79

-9 is missing marital_status

Loneliness Full Model

formula: 
loneliness ~ scale(age) + factor(sex) + scale(SF_12) + relevel(factor(education_state), ref = "3") + relevel(factor(job_sec), ref = "3") + scale(hh_income) + relevel(factor(hh_comp), ref = "3") + relevel(factor(marital_status), ref = "Partnered")
data:    data

Coefficients:
                                                              Estimate Std. Error   z value Pr(>|z|)    
scale(age)                                                  -9.262e-02  7.455e-04  -124.239  < 2e-16 ***
factor(sex)Male                                             -2.076e-01  8.781e-04  -236.396  < 2e-16 ***
scale(SF_12)                                                -8.742e-01  4.711e-04 -1855.889  < 2e-16 ***
relevel(factor(education_state), ref = "3")0                -2.334e-02  1.564e-03   -14.921  < 2e-16 ***
relevel(factor(education_state), ref = "3")1                -5.810e-02  5.157e-03   -11.267  < 2e-16 ***
relevel(factor(education_state), ref = "3")2                -1.098e-01  1.539e-03   -71.307  < 2e-16 ***
relevel(factor(education_state), ref = "3")5                -1.463e-02  1.856e-03    -7.882 3.21e-15 ***
relevel(factor(education_state), ref = "3")6                -1.198e-02  1.579e-03    -7.585 3.33e-14 ***
relevel(factor(education_state), ref = "3")7                -1.334e-02  1.733e-03    -7.700 1.36e-14 ***
relevel(factor(job_sec), ref = "3")1                        -1.642e-01  2.215e-03   -74.131  < 2e-16 ***
relevel(factor(job_sec), ref = "3")2                        -7.349e-06  1.634e-03    -0.004    0.996    
relevel(factor(job_sec), ref = "3")4                         1.170e-01  1.356e-03    86.289  < 2e-16 ***
relevel(factor(job_sec), ref = "3")5                         1.994e-01  1.636e-03   121.906  < 2e-16 ***
relevel(factor(job_sec), ref = "3")6                         1.000e-01  1.823e-03    54.863  < 2e-16 ***
relevel(factor(job_sec), ref = "3")7                         2.575e-01  1.323e-03   194.686  < 2e-16 ***
relevel(factor(job_sec), ref = "3")8                         3.983e-01  1.658e-03   240.209  < 2e-16 ***
scale(hh_income)                                            -3.572e-02  4.422e-04   -80.782  < 2e-16 ***
relevel(factor(hh_comp), ref = "3")1                         4.517e-01  1.385e-03   326.003  < 2e-16 ***
relevel(factor(hh_comp), ref = "3")2                         4.652e-01  2.507e-03   185.550  < 2e-16 ***
relevel(factor(hh_comp), ref = "3")4                        -5.316e-02  1.011e-03   -52.608  < 2e-16 ***
relevel(factor(marital_status), ref = "Partnered")Separated  5.609e-01  1.665e-03   336.957  < 2e-16 ***
relevel(factor(marital_status), ref = "Partnered")Single     4.877e-01  1.210e-03   403.041  < 2e-16 ***
relevel(factor(marital_status), ref = "Partnered")Widowed    7.780e-01  3.912e-03   198.876  < 2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Threshold coefficients:
    Estimate Std. Error z value
1|2 0.793914   0.001685   471.1
2|3 3.233672   0.001842  1755.6
(10207 observations deleted due to missingness)

Reference factors: education_state -> 3 - A-level or equivalent job_sec -> 3 - Lower management and professional hh_comp -> 3 - Multiple adults no kids marital_status -> Partnered

Looks about right. Widowed and Separated show more loneliness than Single, and Married show lowest. Single adult households more lonely than multiple adults, but adults with kids show lowest. Moving up the list of job_sec (will paste breakdown below) show increased loneliness. Age and Gender also show relationship we would expect.

Values for job_sec:
1 - Large employers and higher management
2 - Higher professional
3 - Lower management and professional
4 - Intermediate
5 - Small employers and own account
6 - Lower supervisory and technical
7 - Semi-routine
8 - Routine

Minimal Model (just curious)

formula: loneliness ~ scale(hh_income)
data:    data

Coefficients:
                   Estimate Std. Error z value Pr(>|z|)    
scale(hh_income) -0.2025969  0.0003893  -520.4   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Threshold coefficients:
     Estimate Std. Error z value
1|2 0.3863648  0.0003101    1246
2|3 2.3916726  0.0005460    4380
(972 observations deleted due to missingness)
ld-archer commented 1 year ago

Smoking (ncigs)

Ethnicity

Justification for using ethnicity to predict smoking: Bhopal et al.. This review found differences in cigarette consumption between many ethnic minority groups.

NS-SEC

Justification for job_sec (NSSEC) to predict smoking: Hiscock et al. (2014). Paper links socioeconomic status and smoking in England. NS-SEC was 1 of 7 measures of SES used to classify individuals (routine or manual occupation was +1 in an SES score from 0-7).

Also Aspinall and Mitton (2014)

There was a clear social class (NS-SEC) gradient in smoking prevalence for ‘White British’ and ‘Other White’ males and females (Fig. 1). There was also a gradient, less regular partly because of small numbers, in the ‘White and Black Caribbean’ and ‘White and Black African’ groups. This was much less perceptible in the ‘White and Asian’ group. In the Indian, Pakistani and Bangladeshi groups, the gradient was either much more muted or entirely absent. There was some evidence of a gradient in the black groups but a much stronger gradient in the Chinese group, commensurate with that seen in the White groups.

Household Income

Nyakutsikwa, Britton, and Langley (2020)

ld-archer commented 1 year ago

SF-12 MCS

Measure of Physical Health

To generate a measure of physical health, we will use questions from the SF-12 questionnaire that only relate to physical health. These are:

  1. scsf2a : physical health limits moderate activities
  2. scsf2b : physical health limits several flights of stairs
  3. scsf3a : physical health limits amount of work
  4. scsf3b : physical health limits kind of work
  5. scsf5 : pain interfered with work

Some of these variables are scored in the same way - a 5 level scale where 1 is worst and 5 is best. e.g. Physical health limits amount of work:

  1. All of the time
  2. Most of the time
  3. Some of the time
  4. A little of the time
  5. None of the time

Some are scored slightly differently but in the same 'direction' for want of a better word. e.g. Health limits several flights of stairs:

  1. Yes, limited a lot
  2. Yes, limited a little
  3. No, not limited at all

However, one question is the opposite direction (1 - best, 5 - worst). Pain interfered with work:

  1. Not at all
  2. A little bit
  3. Moderately
  4. Quite a bit
  5. Extremely

These answers will need to be flipped.

We can create a continuous variable for physical health from these answers, where lower values equal better physical health. This will then be included in the SF-12 MCS model.

The physical health score phealth is a mean summary score of these 5 questions, which ranges from 1-5 (including -9 as missing).

physical_health_score

There are 1146/~32000 missing in wave 11.

Full Model (with lagged SF-12)

Call:
lm(formula = formula, data = data, weights = weight)

Weighted Residuals:
       Min         1Q     Median         3Q        Max 
-3.974e-11 -1.500e-13  0.000e+00  9.000e-14  9.306e-10 

Coefficients:
                                               Estimate Std. Error    t value Pr(>|t|)    
(Intercept)                                   4.771e+01  4.542e-15  1.050e+16  < 2e-16 ***
scale(age)                                    1.311e-14  2.645e-15  4.957e+00 7.28e-07 ***
factor(sex)Male                              -6.846e-15  3.797e-15 -1.803e+00  0.07144 .  
relevel(factor(ethnicity), ref = "WBI")BAN   -9.422e-15  2.471e-14 -3.810e-01  0.70295    
relevel(factor(ethnicity), ref = "WBI")BLA    9.292e-15  1.470e-14  6.320e-01  0.52725    
relevel(factor(ethnicity), ref = "WBI")BLC    8.376e-15  2.088e-14  4.010e-01  0.68836    
relevel(factor(ethnicity), ref = "WBI")CHI   -1.856e-14  2.515e-14 -7.380e-01  0.46065    
relevel(factor(ethnicity), ref = "WBI")IND   -1.062e-15  1.114e-14 -9.500e-02  0.92410    
relevel(factor(ethnicity), ref = "WBI")MIX   -7.133e-15  1.354e-14 -5.270e-01  0.59826    
relevel(factor(ethnicity), ref = "WBI")OAS    1.916e-14  1.596e-14  1.200e+00  0.22998    
relevel(factor(ethnicity), ref = "WBI")OBL    7.465e-15  6.621e-14  1.130e-01  0.91022    
relevel(factor(ethnicity), ref = "WBI")OTH   -1.479e-14  2.809e-14 -5.270e-01  0.59849    
relevel(factor(ethnicity), ref = "WBI")PAK   -2.121e-15  1.672e-14 -1.270e-01  0.89903    
relevel(factor(ethnicity), ref = "WBI")WHO    9.181e-16  7.194e-15  1.280e-01  0.89845    
scale(hh_income)                              4.243e-15  1.809e-15  2.346e+00  0.01900 *  
scale(SF_12)                                  1.067e+01  1.983e-15  5.381e+15  < 2e-16 ***
relevel(factor(housing_quality), ref = "1")0 -4.214e-15  5.780e-14 -7.300e-02  0.94189    
relevel(factor(housing_quality), ref = "1")2 -2.673e-15  3.937e-15 -6.790e-01  0.49724    
relevel(factor(housing_quality), ref = "1")3 -3.277e-15  7.031e-15 -4.660e-01  0.64119    
relevel(factor(job_sec), ref = "3")1          2.672e-14  9.182e-15  2.910e+00  0.00363 ** 
relevel(factor(job_sec), ref = "3")2          1.245e-16  6.940e-15  1.800e-02  0.98569    
relevel(factor(job_sec), ref = "3")4          1.950e-16  5.902e-15  3.300e-02  0.97364    
relevel(factor(job_sec), ref = "3")5          7.644e-16  6.994e-15  1.090e-01  0.91296    
relevel(factor(job_sec), ref = "3")6          6.960e-16  7.696e-15  9.000e-02  0.92794    
relevel(factor(job_sec), ref = "3")7         -7.968e-16  5.638e-15 -1.410e-01  0.88761    
relevel(factor(job_sec), ref = "3")8          7.466e-16  7.054e-15  1.060e-01  0.91571    
scale(phealth)                               -2.990e-15  2.572e-15 -1.162e+00  0.24514    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 9.407e-12 on 9901 degrees of freedom
  (10118 observations deleted due to missingness)
Multiple R-squared:      1, Adjusted R-squared:      1 
F-statistic: 1.264e+30 on 26 and 9901 DF,  p-value: < 2.2e-16

Without lagged SF-12

Call:
lm(formula = formula, data = data, weights = weight)

Weighted Residuals:
    Min      1Q  Median      3Q     Max 
-4859.3  -108.0     0.0   238.7  2620.6 

Coefficients:
                                             Estimate Std. Error t value Pr(>|t|)    
(Intercept)                                  46.84309    0.24545 190.847  < 2e-16 ***
scale(age)                                    3.25027    0.13927  23.337  < 2e-16 ***
factor(sex)Male                               1.87405    0.20448   9.165  < 2e-16 ***
relevel(factor(ethnicity), ref = "WBI")BAN   -1.61794    1.33596  -1.211 0.225896    
relevel(factor(ethnicity), ref = "WBI")BLA    2.18292    0.79442   2.748 0.006010 ** 
relevel(factor(ethnicity), ref = "WBI")BLC    1.16233    1.12920   1.029 0.303342    
relevel(factor(ethnicity), ref = "WBI")CHI   -1.83940    1.36013  -1.352 0.176287    
relevel(factor(ethnicity), ref = "WBI")IND    2.22519    0.60225   3.695 0.000221 ***
relevel(factor(ethnicity), ref = "WBI")MIX    0.23853    0.73206   0.326 0.744558    
relevel(factor(ethnicity), ref = "WBI")OAS    3.57307    0.86244   4.143 3.46e-05 ***
relevel(factor(ethnicity), ref = "WBI")OBL    8.23099    3.57929   2.300 0.021491 *  
relevel(factor(ethnicity), ref = "WBI")OTH    0.28568    1.51909   0.188 0.850831    
relevel(factor(ethnicity), ref = "WBI")PAK    2.22223    0.90370   2.459 0.013948 *  
relevel(factor(ethnicity), ref = "WBI")WHO    2.14112    0.38842   5.512 3.63e-08 ***
scale(hh_income)                              0.32148    0.09776   3.288 0.001011 ** 
relevel(factor(housing_quality), ref = "1")0  4.97712    3.12529   1.593 0.111298    
relevel(factor(housing_quality), ref = "1")2 -0.60156    0.21281  -2.827 0.004711 ** 
relevel(factor(housing_quality), ref = "1")3 -3.26182    0.37880  -8.611  < 2e-16 ***
relevel(factor(job_sec), ref = "3")1          0.53102    0.49652   1.069 0.284872    
relevel(factor(job_sec), ref = "3")2         -1.30004    0.37505  -3.466 0.000530 ***
relevel(factor(job_sec), ref = "3")4         -0.37125    0.31917  -1.163 0.244781    
relevel(factor(job_sec), ref = "3")5         -0.16691    0.37818  -0.441 0.658974    
relevel(factor(job_sec), ref = "3")6          1.36673    0.41596   3.286 0.001021 ** 
relevel(factor(job_sec), ref = "3")7          0.63178    0.30480   2.073 0.038220 *  
relevel(factor(job_sec), ref = "3")8          0.51866    0.38140   1.360 0.173901    
scale(phealth)                                3.06351    0.13564  22.585  < 2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 508.7 on 9902 degrees of freedom
  (10118 observations deleted due to missingness)
Multiple R-squared:  0.1192,    Adjusted R-squared:  0.117 
F-statistic: 53.62 on 25 and 9902 DF,  p-value: < 2.2e-16

When lagged SF-12 is included in the prediction of next SF-12, we see that it dominates the model and is pretty much the sole predictor of next state. When we remove it however, many more variables become significant and even the signs change in some cases. This is something to talk about.

ld-archer commented 1 year ago

Lagged SF-12 model

ld-archer commented 1 year ago

Neighbourhood Safety

Refactored the composite to be more balanced, see below for details:

For each of the seven crime variables, combine ‘very common’ and ‘fairly common’ to create a composite ‘fairly or very common’ (these are the small number categories). Then bin responses like this:

  1. Response to all crime questions is “not at all common” (very safe neighbourhood). Justification is that if you perceive no threat at all this is the best possible state.
  2. Responds to 1+ question as “not very common” but no responses to ‘fairly or very common’ (safe neighbourhood). Justification is that on the whole these people probably feel safe but not all is perfect, so probably not quite as desirable as group 1.
  3. Responds to 1+ question as fairly or very common’ (not safe). Justification is that if perception of crime is very or fairly common, no matter what category, you are likely to feel that your neighbourhood safety is compromised.
crime_var_list = ['burglaries', 'car_crime', 'drunks', 'muggings', 'racial_abuse','teenagers', 'vandalism']

Full Model

formula: 
neighbourhood_safety ~ scale(age) + factor(sex) + relevel(factor(job_sec), ref = "3") + relevel(factor(ethnicity), ref = "WBI") + scale(hh_income) + relevel(factor(housing_quality), ref = "3") + relevel(factor(region), ref = "South East")
data:    data

Coefficients:
                                                                      Estimate Std. Error  z value Pr(>|z|)    
scale(age)                                                           0.1911170  0.0005415  352.918   <2e-16 ***
factor(sex)Male                                                      0.0992345  0.0008171  121.443   <2e-16 ***
relevel(factor(job_sec), ref = "3")1                                 0.0907836  0.0019143   47.425   <2e-16 ***
relevel(factor(job_sec), ref = "3")2                                -0.0183678  0.0014909  -12.320   <2e-16 ***
relevel(factor(job_sec), ref = "3")4                                -0.1669641  0.0013137 -127.095   <2e-16 ***
relevel(factor(job_sec), ref = "3")5                                -0.0022353  0.0014955   -1.495    0.135    
relevel(factor(job_sec), ref = "3")6                                -0.2333180  0.0017010 -137.169   <2e-16 ***
relevel(factor(job_sec), ref = "3")7                                -0.1715113  0.0012251 -139.996   <2e-16 ***
relevel(factor(job_sec), ref = "3")8                                -0.3576329  0.0015141 -236.198   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")BAN                          -0.2870159  0.0058240  -49.281   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")BLA                           0.4235247  0.0038038  111.343   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")BLC                           0.1018040  0.0046766   21.769   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")CHI                          -0.1630809  0.0051622  -31.591   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")IND                           0.1195040  0.0026874   44.468   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")MIX                          -0.0850860  0.0034566  -24.616   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")OAS                           0.2912039  0.0037542   77.568   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")OBL                          -0.8949582  0.0154625  -57.879   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")OTH                           0.4904773  0.0053127   92.322   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")PAK                          -0.2786344  0.0038326  -72.701   <2e-16 ***
relevel(factor(ethnicity), ref = "WBI")WHO                           0.1278501  0.0017688   72.282   <2e-16 ***
scale(hh_income)                                                     0.0574468  0.0004580  125.424   <2e-16 ***
relevel(factor(housing_quality), ref = "3")0                        -1.3560440  0.0139548  -97.174   <2e-16 ***
relevel(factor(housing_quality), ref = "3")1                         0.3235484  0.0016804  192.543   <2e-16 ***
relevel(factor(housing_quality), ref = "3")2                         0.1504321  0.0016288   92.360   <2e-16 ***
relevel(factor(region), ref = "South East")East Midlands            -0.1127611  0.0017424  -64.715   <2e-16 ***
relevel(factor(region), ref = "South East")East of England          -0.0833717  0.0016068  -51.887   <2e-16 ***
relevel(factor(region), ref = "South East")London                   -0.9388216  0.0015856 -592.098   <2e-16 ***
relevel(factor(region), ref = "South East")North East               -0.2331798  0.0022402 -104.087   <2e-16 ***
relevel(factor(region), ref = "South East")North West               -0.3220310  0.0015884 -202.739   <2e-16 ***
relevel(factor(region), ref = "South East")Northern Ireland          1.2223204  0.0031364  389.717   <2e-16 ***
relevel(factor(region), ref = "South East")Scotland                  0.1814973  0.0018498   98.119   <2e-16 ***
relevel(factor(region), ref = "South East")South West                0.3125546  0.0016802  186.023   <2e-16 ***
relevel(factor(region), ref = "South East")Wales                     0.3207328  0.0023706  135.295   <2e-16 ***
relevel(factor(region), ref = "South East")West Midlands            -0.2283518  0.0017088 -133.630   <2e-16 ***
relevel(factor(region), ref = "South East")Yorkshire and The Humber -0.2783216  0.0016944 -164.264   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Threshold coefficients:
     Estimate Std. Error z value
1|2 -0.836559   0.001999  -418.5
2|3  1.437975   0.002016   713.2
(9429 observations deleted due to missingness)
ld-archer commented 1 year ago

Neighbourhood Safety cont.

Counts of new composite. neighbourhood_safety_sounts

RobertClay commented 1 year ago

looks good. happy to integrate this into new docs. can import Rmd directly with refs/figures. want to dicuss?

ld-archer commented 1 year ago

Closing as completed on branch 84, then brought into branch 113 with lots of other changes. Development branch was then created directly from branch 113, so it has been in development from the start.