On using first differences, this is probably the key reason I generally use monthly data, as the monthly data contains the average of more than one poll.

So while using monthly average data would substantially reduce the oscillating nature of the series, it doesnt completely remove it all of the time.This is just one of a number of small problems in trying to do any empirical analysis on polling series data that we arent sure is true, but we know is pretty close to being true. So we really have to make the assumption that the average monthly series is a true representation of political opinion in that month, even though it probably isnt exactly true, but also while aknowledging that it is far more likely to be true than any individual poll in that month.

The monthly average series is the reason why the DW is a little over 2 not a lot higher as you would expect, because the negative correlation has been substantially reduced by the averaging process.That little over 2 value is quite a reoccuring DW value in polling regressions.

But as you say, in reality it just underestimates significance. I can live with that – actually I have no choice but to live with it It’s far better than the reverse.

On testing for the cumulative nature of rate rises, unfortunately the effects over time get washed out by other events, so it becomes impossible to separate out the rate components part from other issues.

With the ANOVA stuff – the big problem there is with the best specced model.The best specced model for the ALP vote uses PM dissatisfaction as a key variable. Yet the cash rate appears to seriously contribute to that as well – but since the PMdissat and cashrate are correlated over time the whole thing becomes an exercise in blech… if you get my drift!

Unfortunately there is no actual set of statistical processes that work perfectly with polling data because the data itself contains so much uncertainty, its inertial and its very very laggy. So the best I can do (but am certainly open to suggestions here) is to play around the edges with it and make a few assumptions about the data to be able to pull out fairly accurate relationships, even if not as accurate as one could achieve with more ordinary types of data like basic economic data.

BTW: As a matter of interest the T-statistic of 2.52 from the regression is just the ordinary t-test statistic for comparing the four months where there was a interest change to the others when there wasn’t.

So bottom line is this doesn’t matter for your conclusion. What matters much more is that you have gone on a “fishing expedition” trying to discover the relationship (if any) between ALP vote and IR changes. Regressing on this months change doesn’t work. So I will try last month. Why not the month before, or the month before that? The pain of an Increase in rates will tend to accumulate over time. It might be 6 months before someone in mortgage stress really starts to feel the difference and blame JH.

In assessing the significance of the lag-1 terms that you identified, you need to look at a model that includes all terms you could reasonably consider and then look at the overall ANOVA F-statistic. I suspect that the P-value will not be nearly as impressive as 0.0068.

Anyway, I am being a picky dickhead! I am sure you have a day job and don’t have unlimited time to do a perfect statistical analysis on the fly. I think it is great that you are bringing some empirical discipline to the political discussion.

It might be interesting to look at how the variable “time since last interest rate increase” affects TPP.

Re Howard’s Penrith shopping centre chivalry and the manner in which HE has decided to let faithful Party acolytes and stooges fend for themselves outside The Firewall in the last desperate days of The Rat Empire. The writing’s from Bill Burroughs but its relevence echoes with Nicholson’s animated cartoons in Rag Rupert, as well as the men women and children who were “watchfully abandoned” when SIEV-X disappeared from the radar blips of RAN vessels under Howard’s orders.

“Somewhere in the shadow of the appalling Titanic disaster
slinks–still living by the inexplicable grace of God–a cur
in human shape, to-day the most despicable human being in
all the world.

In that grim midnight hour, already great in history, he
found himself hemmed in by the band of heroes whose watchword
and countersign rang out across the deep–”Women
and children first!”

What did he do? He scuttled to the stateroom deck, put
on a woman’s skirt, a woman’s hat and a woman’s veil, and
picking his crafty way back among the brave and chivalric
men who guarded the rail of the doomed ship, he filched a
seat in one of the life-boats and saved his skin.”

Hey, I promise I’ll get it re-graded too, if things turn out right. That rain the other day created some new furrows, but the sun seems to be shining again now!

Cheers

Rod

“The bloggers who have perhaps been most important and prominent down under are psephologists – specialist electoral statisticians who try to understand and analyse polls, and consider the interlocking numbers games of electoral politics.

Head counters like the anonymous Possum Comitatus, Simon Jackman, William Bowe, and Peter Brent produce accessible, incisive, original takes on polling, and engage in prolonged discussion with their readers about the meaning and import of their analysis.” ABC Online.

“Most important and prominent”, eh?
Well done, son! Bloody well done. Naturellement, we already knew.

Yep, wouldn’t be surprised if the 5pm news stations run with this tonight.

see http://www.theaustralian.news.com.au/story/0,25197,22729946-12377,00.html

Cheers

Rod