I’m having a hard time listening to talking heads even on TA3 talking about exit poll numbers as if they are real, when they are often quite far off. There’s no attempt to say, hey, these polls were off systematically last time. Instead, there’s blather by otherwise smart people (see the bottom of this post). What if we adjusted current results by what happened last time?
First MVK:
| MVK | Exit poll | Seats | Adjustment based on last time | adj. seats | |
| smer | 37.3 | 69 | 6.79 | 44.09 | 83 |
| kdh | 10.8 | 20 | -1.38 | 9.42 | 18 |
| olano | 7.6 | 14 | ? | 7.6 | 14 |
| sdku | 7.5 | 14 | -0.38 | 7.12 | 13 |
| Most-Hid | 6.8 | 13 | -0.08 | 6.72 | 12 |
| SaS | 5.9 | 11 | -0.26 | 5.64 | 10 |
| SMK | 5.1 | 9 | -1.47 | 3.63 | 0 |
Then FOCUS:
| FOCUS | Exit Poll | Seats | adjustment | adj. share | adj. seats |
| smer | 39.6 | 75 | 5.1 | 44.7 | 80 |
| kdh | 9.9 | 19 | -0.6 | 9.3 | 17 |
| olano | 8.8 | 16 | 8.8 | 16 | |
| sdku | 8.1 | 15 | -2.7 | 5.4 | 9 |
| Most-Hid | 6.3 | 12 | 1.4 | 7.7 | 14 |
| SaS | 7.1 | 13 | 0.5 | 7.6 | 14 |
| SMK | 4.4 | 0 | -2.0 | 2.4 | 0 |
| SNS | 4.1 | 0 | -1.0 | 3.1 | 0 |
So what to read from this? Better chance than might be apparent how for a one-party Smer government (or a Smer-coalition with an expendible partner).
Recaption: "All election-night political commentary"
| MVK | Exit poll | Seats | Adjustment based on last time | adj. seats | |
| smer | 37.3 | 69 | 6.79 | 44.09 | 83 |
| kdh | 10.8 | 20 | -1.38 | 9.42 | 18 |
| olano | 7.6 | 14 | ? | 7.6 | 14 |
| sdku | 7.5 | 14 | -0.38 | 7.12 | 13 |
| Most-Hid | 6.8 | 13 | -0.08 | 6.72 | 12 |
| SaS | 5.9 | 11 | -0.26 | 5.64 | 10 |
| SMK | 5.1 | 9 | -1.47 | 3.63 | 0 |
good job, kevin, appreciate your work
Kevin, a brilliant idea, but a very strong assumption as well. Can we assume even the same direction, if not the effect, of bias?
No you are absolutely right–we cannot make those kind of assumptions and so what I offer is merely food for thought. There were /some/ patterns between 2006 and 2010. My main point was to look at the degree of doubt that we should use when dealing with exit polls… But my guess is that there may be /some/ underestimation of Smer…
Thanks Kevin,
great job as usual. But have some compassion with those poor people. What would they talk about for another two hours? This way they have at least two sets of random numbers to macerate.
Too true. Alas, neither STV nor TA3 could show reruns of Panelak in the intervening moments…
Kevin, great idea, adjusted numbers from Focus suggest stronger position of Smer, but this topic is difficult to discuss ina TV discussion… :)
Marek
Why are you correcting for an additive constant, why not use multipliers? And why not model both the offset and slope of the error curve or even some higher order parameters? Furthermore you only consider mean values and discard standard deviation and other information about the distribution. For example, the uncertainly is higher for results of parties with smaller vote share (the CIs are wider here), hence you may want to make stronger corrections to the latter parties’ percentage points.
The idea of making correction is nice, but you need more than elementary school math to do it properly. Otherwise, you may as well use tarot pack readings to obtain your correction…
Hi Kevin, the problem of adjustment of Focus number is that we did not carry out exit poll in 2010. It was only telephone interviewing. But in generally adjustment mainly for SMER is correct in my opinion.