|
Tuesday, 05 April 2005 21:40 |
Written by Graham Young
|
Newspoll's latest poll is of 1145 people and puts its accuracy at plus or minus 3%. So when it shows a swing towards Labor of 3.8% one should be cautious. However, there are grounds for thinking there might be something in this poll.
The result is in line with the Morgan poll that I reported on a few days ago. It is more or less permissible to combine the two samples, which gives us 3,036 respondents (1891 from Morgan and 1145 from Newspoll). Except the sample isn't quite that large as Newspoll excluded 8% of their sample because they didn't answer this question with a preference. Adjusting for this (and assuming Morgan didn't do the same and didn't tell us) the total is 2,944. Still pretty large. The adjusted totals are 44% Coalition, 40% Labor, 7% Greens and 10% other with a sample error of plus or minus 1.8%. In other words this is a statistically meaningful move since the last election.
Newspoll also tends to confirm the results in its own terms because of the trend. Unlike Morgan, where the results have bounced in both directions, Newspoll shows a clear trend since the election. This trend is important because just as I can combine results from different surveys that occur over more or less the same time period, you can do the same with different samples within the same survey over a time period. By producing a rolling average (which is all a trend actually is) you can get a more certain result.
ABC news reports Opposition Treasury Spokesman Wayne Swan on the result:
"Labor treasury spokesman Wayne Swan says the result is welcome but it is a long time before the next election.
He says the poll reflects the public's disappointment in the recent interest rates rise.
He's right to be cautious about the poll result, and politic on the reasons for the change. While we can measure movement on these figures we can't measure cause. That's why What the people want asks qualitative questions and combines them with regression analysis based on the ABS statistics, wherever possible.
|
|
Saturday, 02 April 2005 01:19 |
Written by Graham Young
|
|
The latest Roy Morgan poll of federal voting intentions is out. You can read it at www.roymorgan.com/news/polls/2005/3848/. It shows the Government on 44%, 2.4% below their result at the last election, and the ALP on 39%, 1.4% higher.
Gary Morgan spends some lines analysing the movement and concludes: "This latest Morgan Poll shows the Primary vote has stabilised after the reaction against interest rates in early March. Primary support for the ALP has risen and an election at this time would be close."
In fact it shows nothing of the sort. At the bottom of the page Morgan has a table showing the margin of error inherent in samples of 1,000. At 1,891 the sample on which this analysis is based is much larger than 1,000, so this table doesn't help. However, I've found a web page that not only explains the concept of margin of error, but has a calculator where you can work it out for yourself. In the case of Morgan's sample, his margin of error for the Liberal Party vote is 2.2% in 95% of cases. In other words, the Liberal Party vote could be anywhere between 41.8% and 46.2%, meaning that on the basis of this poll it could have either lost an election decisively to Labor, or won it with a similar margin to last year - take your pick.
We pollsters spend a lot of time making up reasons for why things appear to be happening, when nothing is really happening at all. This appears to be yet another case of this phenomenon.
|
|
<< Start < Prev 31 32 33 34 Next > End >>
|
|
Page 34 of 34 |
