I was inspired by this post at The Monkey Cage to repeat an exercise I’ve previously carried out for Italy: namely, to calculate the expected duration of the new cabinet using some off-the-shelf models. I think it’ll last a full five year term — here’s why.
I grabbed some recent data about coalition duration from Matt Golder’s site.
I then carried out a simple survival analysis using a number of independent variables:
- pre-electoral coalition or not;
- majority status
- single party status
- government ideological range
- caretaker government or not
- number of formation attempts
- effective number of parliamentary parties
- formal investiture requirement
- and post-election formation or not
which gives you the following model:
Name |
Value |
Std. Error |
z |
p |
| Intercept | ~6.52 | 0.32 | 20.66 | 0.00 |
| Pre-electoral coalition | -0.19 | 0.23 | -0.80 | 0.42 |
| Majority govt | ~0.51 | 0.15 | ~3.42 | 0.00 |
| Post-election | ~1.03 | 0.13 | ~7.83 | 0.00 |
| Single party | -0.03 | 0.18 | -0.14 | 0.89 |
| Ideological range | ~0.00 | 0.00 | -0.41 | 0.68 |
| Caretaker | -0.90 | 0.28 | -3.18 | 0.00 |
| Formation attempts | ~0.00 | 0.06 | -0.03 | 0.97 |
| ENPP | -0.15 | 0.06 | -2.71 | 0.01 |
| Investiture | -0.13 | 0.12 | -1.10 | 0.27 |
Plug in values of 2.58 for the effective number of parliamentary parties, and 11.4 on the ideological range (admittedly using figures taken from the 2005 election), and all the other relevant dummies, and you get the following:
- Expected duration (days): 2103
- 10th percentile: 222 days
- 90th percentile: 4843 days
Okay, so there’s a huge amount of uncertainty surrounding the prediction, but still, I’d say that on this basis, if you believe the coalition will fall before 2015, you have to make some kind of argument to the effect that British politics is uncongenial to coalition government.
R code:
library(foreign)
library(survival)
coal<-read.dta("coalition1.dta") ## from matt golder's site
my.exp.basic<-survreg(Surv(duration,ciep365==0)~ pec+majority2+postelection+singleparty+range+caretaker+ incbargain+enpp2+investiture,data=coal, dist="exponential")
predict(my.exp.basic, newdata= data.frame(pec=0,majority2=1,singleparty=0,range=11.4, caretaker=0,incbargain=1,enpp2=2.58,investiture=0,postelection=1), type="response")
predict(my.exp.basic, newdata= data.frame(pec=0,majority2=1,singleparty=0,range=11.4,caretaker=0,incbargain=1,enpp2=2.58,investiture=0,postelection=1), type="quantile", p=c(0.1,0.9)))
Could you run that one for a Lib-Lab coalition, just out of curiosity?
Link | May 16th, 2010 at 12:16 pm
Yep. If you run it for a LibLab minority government, then you plug in the following values:
range = 6
incbargain=2 (because it would have been after the ConLib attempt)
majority2 = 0 (because it’s not rainbow)
And you get an expected duration of 1273 days, or three and a half years, with 10th percentile of 134 days, and a 90th percentile of 2930 days.
If you assume that it’s a rainbow coalition with a majority, then you get almost the same results as the LibCon coalition, largely because the model can’t take account of range on other dimensions (centre-periphery). That’s a clear limit of abstraction.
Link | May 16th, 2010 at 12:40 pm
It’s too late and I’m too lazy to research this myself, so briefly, is the model affected by the voting system? Would it be possible too, for you to show the complete distribution?
TIA
Nice work
Link | May 17th, 2010 at 1:20 am
Hi Ron,
The model isn’t directly affected by the voting system. It’s affected indirectly because the voting system changes the effective number of parliamentary parties. If AV were introduced, the ENPP would likely go up very very slightly; if other more proportional systems were introduced, if would go up more.
For the distribution — yeah, I’ll need to poke around a bit before I can plot that.
Link | May 17th, 2010 at 1:25 pm