Monday, March 03, 2008

Not meaningful

On her blog, Audrey Young has a nice analysis of the Herald's latest poll, including a look at voter's coalition preferences. Unsurprisingly, this finds that National voters would prefer their party to deal with NZ First or United Future, while Labour voters would prefer Labour to finally consummate their relationship with the Greens. Unfortunately, it goes on to report the coalition preferences of subsampled smaller parties as well. To illustrate the problem here, I'll reproduce Young's quote with the appropriate margins of error inserted:

More National supporters backed a deal with New Zealand First, 30.2 per cent [4.9%], over other parties. Next preferred was United Future with 21.8 per cent [4.9%], and even 17.8 per cent [4.9%] of National supporters preferred a deal with the Greens, who are more natural allies of Labour. Only 11.4 per cent [4.9%] of National supporters preferred a deal with the Maori Party.

More Labour supporters backed the Greens, 42 per cent [6.0%], over any other party for Labour to deal with, but 34.6 per cent [6.0%] preferred New Zealand First, Labour's present confidence and supply partner.

Even fewer Labour supporters than National wanted to deal with the Maori Party, 8.8 per cent [6.0%].

New Zealand First supporters overwhelmingly think the party should form a government with National, 90.9 per cent [24.5%], compared with 9.1 per cent who prefer Labour [24.5%].

A large majority of Green Party supporters would prefer to deal with Labour, 69 per cent [17.3%]; but almost a quarter, 24.1 per cent [17.3%], favour a deal with National.

A majority of Maori Party supporters would prefer to deal with Labour rather than National, but the split is less pronounced: 57.1 per cent [29.5%] favour Labour and 42.9 per cent favour National [29.5%].

All of Act [56.6%] and United Future's [56.6%] supporters would prefer a deal with National.

(Margins of error calculated based on the backcalculated subsample size, using the 0.98/SQRT(n) rule of thumb)

Once you have this information, it becomes apparent that much of this section of Young's report is simply not statistically meaningful. In some cases, it's even laughable - the ACT and United Future "results" are based on the opinions of three people. Now, if you want to report the opinions of three random people, fine - but you shouldn't call it a poll.

Coalition preference data is important. It provides valuable information to voters about what sort of government they might get from certain choices, and valuable information to politicians about what their voters want them to do. And I'd very much like to see some robust data in this area. But this just isn't it, and I very much doubt any media organisation would be willing to go to the effort of collecting it.