Popular Posts

Caveat Emptor

The opinions expressed on this page are mine alone. Any similarities to the views of my employer are completely coincidental.

Friday 3 July 2015

A little light in a murky place

This is a little teaser: more will follow in due course, probably in the late Autumn.

One of the things that irritated  me about the GBCS was the seemingly endless delay in making the data available for replication. Well, eventually it emerged, and as I've remarked before, the documentation  is quite an eye opener. There are a few unfortunate typos in it which may throw the less dedicated off the scent, but with a bit of perseverance and a clue from someone in the know (thanks, you know who you are) I've been able to reconstruct almost exactly what they did (the main problem was in trying to guess the exact specification of the model they fitted - what seems to those at the theoretical heights like petty detail not worth mentioning  is actually quite important to us knuckle dragging under-labourers).

I'm not going to tell you everything I've discovered right now. For the moment I'm going to take the GBCS  7 class model at face value (let's not think about the missing 8th class -  I like to call them the New Socialites). I'm not going to make any remarks about the logic of the inductive procedure; I'm simply going to accept the outcome of their data mining model fitting and say: OK, if you believe in that, then, on the basis of your own data you must accept the following implications.

I've ignored the GBCS data itself and what follows is based entirely on the weighted GfK survey data which I've treated 'as if' it is a probability sample. Disregarding the GBCS data has no consequences since it contributes no information to the parameter estimation. In examining the relationship between the GBCS classes and other variables I've used the so called three-step method (ML for nominal variables, BCH for continuous variables). This is important  because the modal category assignment procedure used in most (all?) of the Savage et al. papers seriously underestimates the relationship between the classes and external variables.

The first thing I'm curious about is how the GBCS classes are related to the NS-SECs. The latter play something of the role of the albatross to the GBCS ancient mariner. They are always there, lurking in the background, and the mariner has an itchy trigger finger. However, so far I've not discovered anything in the public domain that displays the relationship. So here we go (click on the pictures to make them readable):


Now I'm not so naive as to think that I could get away with this. I can just hear the chuckles in the Boulevard St Houghton: oh dear, what a quaint piece of anglo-empiricism!

 What we need, of course, is to bring out the full 'relational' nature of the dinks-bumps with a bit of correspondence analyses, so here is a CA bi-plot of the same data.

What it shows is that the GBCS classes and the NS-SECs are clearly empirically related to each other but that there are two groups of GBCS classes which are pretty homogenous in terms of their NS-SEC profiles.

The Established Middle Class (EMC) and the Technical Middle Class (TMC) look much the same in NS-SEC terms, as do the  New Affluent workers (NAW) and the Traditional Middle Class (TMC)

Though the NS-SECs don't distinguish these pairs clearly, other things do, as you'll see in a moment.

Before passing on to that however, one final comment about the NS-SECs and the GBCS classes. Look at the first column of the table. This shows the distribution of the GBCS Elite: two thirds of them are located in the NS-SEC categories Lower Managers and Professionals and "below". This is, to say the least, odd. Think of the implications of this for the sleight of hand worked in the recent Sociological Review Special Issue on the Elite where the top  NS-SEC categories were used interchangeably with the GBCS Elite category.

But let's move on. Next up is a table that contains some information about how the GBCS classes are related to a bunch of other social and demographic stuff. Same estimation strategy as before.





I'll leave you to rummage through this treasure chest pretty much on your own. But before I go I'll just make 4 points:

Point 1. Look at the uncertainty around the estimates of the percentages in each class! I think it really matters whether the "Elite" is 2 % or 12% of the population, likewise whether the "Precariat" is 9% or 19%.

Point 2. The "Traditional Working Class" is old, retired, disabled and female! Live long enough and you'll end up there.

Point 3. The "Emergent Service Workers" are youngsters who are single or living alone in rented accommodation. More than a 10th of them haven't yet finished their education.

Point 4. The "Elite" live in London and the South-East, they are married and our best guess is that at least half of them are female. So much for the glass ceiling. In GBCS land the sisters have broken through. Or is something a bit wrong here?

There is endless fun to be had with these data as the full horror emerges, but that's all for now. See you again around November...




2 comments:

Anonymous said...

Shame your plot has TMC twice...

Colin said...

Er, perhaps you need to read it more carefully. It has TMC once and TWC once. A little fast on the trigger finger...