On March 5, New Zealand held its census. And for four months starting next Monday, 25 March votingage New Zealanders of Māori descent will have the option of switching between the Māori and general electoral rolls. The results of the census, and the Māori Option will be used to draw electorate boundaries for the next two general elections.
My post in late 2011 looking at the legality of this year’s Māori option is – as I predicted and as was predictable – now of academic interest at most. But there is an older – and I suspect, very little known – story of substantial importance to the Māori option. The story of how, prior to the 1996 census, and the boundary redrawing that followed in 1997, the Government Statistician decided to increase the number of Māori seats, because it seemed like a good idea.
Since the introduction of the mixed member proportional voting system (MMP), the number of Māori seats has been determined on the same basis as general seats (within +/ 5%, and a margin created by the need for there to be whole number of seats, the number of residents served by each electorate MP is the same). Before the Electoral Act 1993, the number of Māori seats was limited to four, no matter how many people were registered to vote on the Māori roll. The unfairness that previously existed can most clearly be seen from the very beginning of MMP. Between 1993 and 1996, the number of Māori electorates increased from 4 of 99 electorates (4.04%) to 5 of 65 electorates (7.69%).
Because we have chosen to determine our electorates with respect to the overall ordinarilyresident population (not the enrolledtovote or voting age population), data from both the census, and enrolment forms is used to determine the Māori Electoral Population, which is used to determine the number of Māori seats after each Māori option is concluded. The formula is relatively simple, but understanding it is unnecessary to appreciate what is going on.[1]
The Māori electoral population is the group of people whose electorate representation in Parliament is through the Māori seats. It is determined by taking the Māori descent population (i.e. the number of ordinarilyresident New Zealanders who have a Māori ancestor), and assigning a proportion of this group according to proportion who are enrolled on the Māori roll. So, if 60% of people who are listed on the electoral roll as being of Māori descent are enrolled on the Māori roll, then 60% of the overall Māori descent population counts as the Māori electoral population, and the other 40% are included in the general electoral population.
But how do we know what the Māori Descent population is? Well, question 14 in the Census asks:
Are you descended from a Māori (that is, did you have a Māori birth parent, grandparent or greatgrandparent, etc)?
It is the answer to this question (and not the ethnicity question) that is used in determining the number of electorates of each type.
For example, the calculation of the number of Māori seats for the first MMP election was, in short form:
 On their 1991 census forms, 511,278 people indicated they had Māori ancestry.[2]
 After the 1994 Māori Option, 51.7% of people who indicated on their enrolment forms that they had Māori ancestry were on the Māori roll, so the Māori electoral population was 264,222 (which is ~51.7% of 511,278).
 This Māori electoral population was enough people (not voters) for 5.11 Māori seats, which got rounded down to 5.
A Māori Option is held after each Census, where these numbers get updated. The results of these calculations are simple enough:
Māori Option 
Elections covered 
‘Fair’ number of Māori seats 
Māori seats 
1994 
’96 
5.11 
5 
1997 
’99 
6.23 
6 
2001 
’02 & ’05 
6.84 
7 
2006 
’08 & ’11 
7.25 
7 
The fair number of seats has increased each time, but with rounding, the last increase wasn’t big enough to change the actual number. These increases have been a result of two broad factors: demographic change (with the Māori descent population rising faster than the population overall), and an increase in the proportion of voters of Māori descent choosing to be on the Māori roll (from 51.7% in 1994, up to 57.6% after the 2006 Māori Option).
You might assume that – albeit with different numbers – the same overall calculation would be done after each census and Māori option. And it largely is – but with one large difference.
For the 1997 boundary determination, the Government Statistician decided he’d do things differently. The change in approach has continued since, and that choice – not demographic changes or Māori enrolment choices – was the reason that at the last two boundary determinations, there were seven Māori seats, instead of the six we would otherwise have had for the last 15 years. The change wasn’t preceded by an amendment to the Electoral Act, or the Statistics Act, but was simply something that the Government Statistician decided would be sensible to do differently.
In the calculation for the first set of MMP boundaries, shown in bullet point form above, you may have noted that the number of people of Māori descent in the first bullet point is the same as the number used in the second bullet point, which is the number of people who answered the census question about Māori ancestry[3] by saying that they were of Māori descent.
This didn’t happen when the calculation was performed again after the 1997, 2001 and 2006 Māori options: the number of people who said “I am of Māori descent” wasn’t the number used as the “Number of People of Māori descent” input. Added to this number was a proportion of people who answer the Māori descent question with “I don’t know” or who gave confusing answers (for example, by leaving it blank, or selecting more than one option).
Statistics New Zealand – figuring that some of those people would be of Māori descent, while others wouldn’t – analysed the other census information provided by those people, and assigned each of them a likelihood of being of Māori descent using a statistical technique known as CHAID (Chisquared Automatic Interaction Detection).
CHAID is a technique frequently used in data mining and massmarketing. It calculates the likelihood of an unknown characteristic (in this case “is this person of Māori descent?”, other times it might be “is this person a credit risk?”), based on other known characteristics, and the existence of a large enough data set to match it against (and what’s larger than a census?).
After assessing 12 variables (arising from other census questions) to see how well they correlated to clear Māori descent answers on others’ census forms, the four that were the best indicators of whether someone was of Māori descent were adopted.[4] One additional variable was added, the place of residence (North Island/Chatham Island or South Island), not because of a correlation to Māori descent, but because of its importance to electoral calculations – resulting in 448 subgroups covering all the different combinations of census responses for those five variables.[5] If, of the people who clearly answered the Māori descent question in one of these subgroups, 6% had said they were of Māori descent, then Māori descent would be imputed to 6% of those who answered “don’t know” or gave an unclear answer.[6]
This process of imputing Māori descent to people who give unclear answers to the census question increases the Māori descent population that is used in electoral calculations:

1994 
1997 
2001 
2006 
Number of people who clearly indicated Māori descent 
511,278 
579,714 
604,110 
643,977 
Māori descent figure used in election calculations 
511,278 
628,429 
670,264 
721,431 
Interestingly, the results of this analysis are only used in the calculation of the number of electorates, and aren’t presented alongside the census results. If they were, we might think of the Māori population of New Zealand as more like 18%, rather than the 1516% commonly believed.
The decision to impute Māori descent in this way has a number of effects:
 it increases the Māori descent population (which increases the number of Māori seats);
 it decreases the general electoral population of the South Island (which lowers the South Island quota, making electorates smaller, and increasing the number of Māori seats and the number of general seats in the North Island);
 it decreases the general electoral population of the North Island (decreasing the number of general seats in the North Island: to date, by more that the increase caused by the second factor)
It seems that almost no one (except perhaps readers of back copies of the Australian and New Zealand Journal of Statistics) knows about this decision, and more importantly, I don’t believe anyone has quantified its effect on the number and type of electorates.
As you may have guessed, I now have. A number of helpful people at the Electoral Commission and the Electoral Enrolment Centre, Statistics New Zealand, and even the Parliamentary Library who have responded to my many information requests over the last months have my thanks [Statistics New Zealand in particular, are excellent responders to information requests, often replying within hours (in the past, I’ve had and apology from them when they were going to take more than a day)].
After each Māori option, the Electoral Act requires election authorities[7] to send the end result of the option to the Government Statistician. Disappointingly, neither the Electoral Commission, nor Statistics New Zealand retains the full information for past Māori options (the split between Islands is crucial). For the three boundary determinations where the decision to impute Māori descent has been in effect, authorities were able to construct full data only for 2007. For 1997, I have been able to glean enough information from the report of the Representation Commission, and another paper published in the Australian and New Zealand Journal of Statistics) to be able to reverse engineer the ratio of Māori enrolment for each island. Unfortunately, the information available in respect of the 2001 redrawing lends itself only to an estimate of the crucial proportions of Māori enrolment by island.
So what is the effect of the decision to impute Māori descent to a proportion of those who don’t know, or wouldn’t say? It has increased the number of Māori seats, and caused a slight decrease in the fair number of general seats in the North Island. As at the last boundary determination, it accounted for 0.8 of the 7 Māori seats:
Māori seats 


1997 
2001 
2006 
With imputation 
6.23 
6.83 
7.25 
Without imputation 
5.73 
6.136.16 (estimated) 
6.44 
Increase 
0.49 
0.670.70 
0.80 
The effect on North Island general seats was present, but wasn’t as marked:
North Island general seats 


1997 
2001 
2006 
With imputation 
44.65 
45.99 
46.74 
Without imputation 
45.00 
46.3146.48 (estimated) 
47.24 
Decrease 
0.34 
0.320.49 
0.51 
Because of rounding, imputation hasn’t had a real effect on the number of North Island general seats (although it may be a matter of time), but as noted above, without it, 2001 would not have seen in an increase in the number of Māori seats, and it would have stayed at six in 2006 as well.
The decision of the Government Statistician – which in a close election could affect the balance of power – hasn’t had a lot of press, with the only public mention in the technical notes of the Government Statistician’s calculation, such as this one in 2006:
The Māori descent figure derived from the 2006 Census includes an allocation of a proportion of those people who did not provide a clear ‘yes’ or ‘no’ response to the Māori descent question. The allocation includes a proportion of those people who answered ‘don’t know’, provided a multiple response (for example ‘yes’ and ‘don’t know’), or who did not provide a response at all to the question.
It also raises important questions. Was the decision proper (in a legal sense)? And perhaps most importantly, is it right?
I don’t have a concluded view on this, but there is cause to question it.
From a statistical standpoint, the analysis will certainly give a more accurate figure for the “real” Māori descent population. Assuming that everyone who answers “don’t know”, or who gives a confusing answer is not of Māori descent is statistically unjustifiable: clearly some of those who don’t know will be of Māori descent. The analysis undertaken is statistically sound, but that observation is diminished when one realises the more accurate information is kept out of census releases, leaving less accurate data for everyone else. Only someone with the full data can undertake a CHAID analysis, and more accurate assessments of not just Māori descent, but also a variety of other variables in the census where people may give confusing answers could be beneficial. That other census data isn’t presented in this way suggests that this isn’t the “number of ordinarily resident persons of New Zealand Māori descent as determined by the last periodical census” that the Electoral Act refers to in the definition of the Māori electoral population.
The underlying principle behind including people who can’t vote in our election calculations is that those who aren’t enrolled (mostly because they are too young to vote) are deserving of the same representation as those are able to vote. With a system of overlapping Māori and General electorates, this is somewhat artificial: we effectively assign Māori descent nonvoters between the Māori roll and the general roll based on a choice they are yet to make. And most importantly, it is a choice that those who don’t know whether they have Māori ancestry, or who won’t clearly answer the question, will never get to make. People who cannot or will not give a clear “yes” to the question ‘are you of Māori descent?’ cannot enrol on the Māori roll.
Having made the policy choice to include those unable to enrol or vote in our electoral calculations, it is clearly necessary to assign the fair proportion of those who would have the choice to enrol on the Māori roll to the Māori electoral population. Increasing the Māori electoral population by assigning to it a proportion of a group who are not given that option, and those who would not be given that choice if old enough to vote, is not so obviously defensible.
Most problematic is that the decision to impute Māori descent to people only affects one stage of the calculation: people who have Māori descent statistically imputed to them are added to the total number of people of Māori descent, but are not added to the number of people of Māori descent who are enrolled on the general roll. At that stage, any who are enrolled are counted as voters on the general roll who are not of Māori descent. Imputing Māori descent to the fair proportion of voters (all of whom will be enrolled on the general roll) would, at that stage, reduce the proportion of those of Māori descent on the Māori roll, and reduce the number of Māori electorates.
The decision was made on legal advice, but the law doesn’t say a great deal about whether imputing Māori descent to those who give unclear answers is acceptable. Both options could probably fit within the words “the total number of ordinarily resident persons of New Zealand Māori descent as determined by the last periodical census”, but one may be a better fit than the other.
The question of whether the Government Statistician should have made a decision likely to have this sort of impact isn’t as interesting as it might seem. Given that these are statisticians, a decision not to do this would have been just as important a choice.
The decision appears to have been made without fanfare, or public or political consultation. Although, again, I’m not sure what that would add given the clear political implications of the decision. It may have been nice if the Royal Commission on the Electoral System had looked at this, but this sort of analysis was very much in its infancy in 1986.
Nonetheless, I am a little disconcerted that a single person can make what is potentially a farreaching decision with no real oversight, and almost in private. I wonder if what should happen in situations like this might be for the government body to seek a court High Court declaration on what the law requires or allows. There will not always be a right answer, but there may be a wrong one.