Template talk:2017 NZ election forecasts

Info on source polls

@Lcmortensen: Shall we discuss the source polls here rather than just reverting each other's edits? My motivation for adding that info was to distinguish between RNZ and Stuff's projections, which are based on multiple polls, and 1 & 3 News's ones, which are just a translation of their own latest poll into seats. (On the UK equivalent you linked me to, they all seem to be based on multiple sources of info.) My fear is that it's misleading to put all four projections on an equal footing. Can you think of another way to communicate this info better? I'm actually wondering if the 1 and 3 News ones should be deleted. Elcalebo (talk) 03:52, 28 August 2017 (UTC)

The prose above the table gives all the information that is needed to differentiate between the polls. Adding the list of individual polls they are based on is just superfluous. Lcmortensen (mailbox) 06:12, 28 August 2017 (UTC)

Fair enough. Hopefully you will consider my recent edit an acceptable compromise between our concern for avoiding superfluous info and my concern for distinguishing between single-poll and multi-poll projections.Elcalebo (talk) 05:47, 30 August 2017 (UTC)

Negotiating disagreements

@Lcmortensen: - can we negotiate disagreements on the talk page, instead of just reverting my/other people's changes? I will create two new sections where we can discuss specific areas of disagreement. Elcalebo (talk) 21:20, 5 September 2017 (UTC)

Overall result: Minimum parties for majority, what the media sources say, or what the parties say?

When entering "coalitions" (see Terminology topic below) in the "Overall result (majority)" section, there seems to be confusion and disagreement about what the source for this info is. Is it what the sources say (e.g. whether Corin Dann on 1 News mentions a Labour-Green-NZ First government or just a Labour-NZ First one)? Or is it based on what the parties themselves say (i.e. Labour have a MOU with the Greens so they would seek to include the Greens alongside NZ First even if they didn't strictly need them)? Or is it just based on the minimum number of parties necessary for a majority (i.e. Labour and NZ First because that would be enough with just two parties, even though in practice there could be a Labour-Green-NZ First government or perhaps a Labour-Green-Māori government ... or National-NZ First instead of National-ACT-Māori-NZ First for the same reasons)? Inevitably we have to include some suppositions about which parties are likely to work together, even in a "minimum majority" approach. I would be interested in what others think is the best approach, and I think it's also tied up with the Terminology question below. Although I had previously argued for "Labour-Green-NZ First (69)" rather than "Labour-NZ First (62)", upon further thought my opinion is: We should take a "minimum parties for majority" approach, so long as we clearly articulate that this is what we're doing in the description and the terminology we use (i.e. don't call them "coalitions"). If there are multiple ways to get to the same majority with the same number of parties, we express both alternatives using slashes. (This is what we are currently doing with "National−NZ First−Māori/ACT coalition (61)" - this is also what the news source said).

In the meantime I have edited and moved @Lcmortensen:'s sentence "When determining coalitions, only the parties necessary for a majority (i.e. 61 seats) are listed as coalition parties." Where it is currently situated is misleading - it makes it seem like this is the logic followed by the people who make the projections, when in fact this is what we have decided on this Wiki page (or at least what @Lcmortensen: has decided). And how it is phrased is also misleading, because 61 isn't a majority if there are two overhang seats.Elcalebo (talk) 21:46, 5 September 2017 (UTC)

Terminology: majority/coalition/other?

I removed the word "coalition" from the "Overall result (majority)" section because "the parties supporting the government may not necessarily all be in a formal coalition - there could instead be confidence and supply arrangements." @Lcmortensen: reversed my change because "No need to nitpick - a coalition refers to any government with two or more parties cooperating, even if it isn't a formal one." What do others think about this? My thinking is that this question is tied up with the question of what the "overall result" thing actually is. If we're going to call it a "coalition" I think this impresses upon us more of a responsibility to speculate about what coalitions are likely to actually form. If we're going to call it a "majority" this lends itself more to a "minimum majority" approach. I am personally opposed to the "coalition" terminology for the reasons I originally expressed, and also because (as I say in the above section) if we're going to take a "minimum majority" approach, I think our terminology should match that instead of giving the false impression that we are proposing actual coalitions that may emerge. Elcalebo (talk) 21:46, 5 September 2017 (UTC)

CARE: Citation conflicts =

Beware not to use <ref name="XXX"> in a template (just <ref> instead), because this causes citation conflicts on the Opinion polling for the New Zealand general election, 2017 page. Preferably, we would use but we can't because the pages the template gets used on might not hold the original referenced citation.

Outdated Stuff seat projections

I've temporarily removed the Stuff seat projections, because it was rather outdated (all the way back to 28 July), and Stuff's poll of polls has moved considerably over the past month (https://interactives.stuff.co.nz/NZ-election-home/index.html), although they have not yet released any seat projections since. Given that the 28 July Stuff seat projections do not reflect the current Stuff's poll of polls at all, (Labour has since surged, Green has collapsed, and United Future's Peter Dunne resigned), I can currently see only see solutions:

1. Removal of any seat projections from the table that are older than 1 month - given that polls are so volatile and dynamic, older information can risk being rather misleading, especially since Stuff's poll of polls are actually regularly updated...
2. Calculate seat projections based on Stuff's current polls of polls by including an overhang for ACT, and allocating the 120 seats of Parliament using the party vote of the four major parties, which risks being Original research.

Sleepingstar (talk) 22:28, 7 September 2017 (UTC)

Thanks @Sleepingstar:. I agree. I have been wondering what to do about the very old Stuff projection also - the poll of polls is updated regularly but they only did a seat projection for their poll of polls launch article on July 28. One idea is that we could include the Stuff poll of polls (along with the Radio NZ and Herald ones) as a separate section on the Opinion polling for the New Zealand general election, 2017 page, called "Polling averages" or something. Then we could just include seat forecasts in this section (the four that are currently there and perhaps Stuff again if they release another article with a seat projection). Elcalebo (talk) 04:50, 8 September 2017 (UTC)

Where'd the Newsroom projection numbers come from? =

Kia ora @HeyItsAndrew: Where did you get those seat projection numbers from the Newsroom poll? from what i can tell, the provided link only mentions percentage numbers. Elcalebo (talk) 11:29, 8 September 2017 (UTC)

Hi @Elcalebo: :) I put the numbers into the electoral commissions website and it gave me the seat projection numbers (http://www.elections.org.nz/voting-system/mmp-voting-system/mmp-seat-allocation-calculator) HeyItsAndrew (talk) 20:54, 8 September 2017 (UTC)

Hi @HeyItsAndrew: I've removed this because it hasn't been our practice so far to include projections which we've worked out ourselves from the pollsters' vote percentages. We've only included projections when the poll organisations (or the media organisations that make polls of polls/forecasts) estimate the seat numbers themselves.

We could change this practice and start including projections like the one you've included. As @Sleepingstar: says above, this could be considered Original research. However, an argument could be made that it's legit to do it, especially if we use the proper website like you've done. However, if we did do this for Newsroom, we should also do it for the Roy Morgan and Listener polls and the Stuff poll of polls. Elcalebo (talk) 05:52, 9 September 2017 (UTC)

Thats completely fine and understandable @Elcalebo: HeyItsAndrew (talk) 06:10, 9 September 2017 (UTC)

@Elcalebo and HeyItsAndrew: Indeed, the use of the electoral commission's seat projection calculator to translate party support from published polling data into projected seats sounds rather awfully close to the standards generally permitted on Wikipedia. I guess there is a somewhat similar precedence of using two official sources of data to synthesis a third derivative set of data (e.g. using land area, and population data to calculate population density, when not provided directly through an official government source). It really all depends on how pedantic one is with adhering to the NOR policy, and is open to subjective interpretation. I'm personally very neutral on this matter, and can see arguments both supporting and refuting this. Nevertheless, should we end up deciding to incorporate data from Stuff poll of polls and Listener/ Roy Morgan, I think it's important that we seperate them into two tables: one for sources that collate >1 poll e.g. RNZ, NZ Herald, and Stuff; and one that incorporates only a single poll: Newshub, One new colmar brunton, Roy Morgan and Listener. Make sure that the Electoral COmmission's seat projection calculator is also cited should the resource be used in making such calculations. - Sleepingstar (talk) 06:16, 9 September 2017 (UTC)

Party colour shading

I've changed the colour shading to indicate the senior party of the parliamentary majority, rather than the largest parliamentary party. The main reason being the final Roy Morgan seat projection which has National as the largest party, yet having no viable options of forming a parliamentary majority. Given that the main goal in the MMP environment is to form a parliamentary majority and therefore form government, I thought that the colour shading should reflect this objective rather than focus solely on the largest party. -- Sleepingstar (talk) 11:37, 15 September 2017 (UTC)

NZ Herald projection

@Langholz8: Not going to lie, the Herald projection has the potential to be incredibly confusing because they give independent projections for each party, rather than using a consistent and congruent model for all parties. Nevertheless, I do feel that the space provided in the editing line is unlikely to provide sufficient room to explain the rationale behind your edits. I still have the same question after these few rounds of editing:

• It appears that your preference is to take 120 as the total seats for calculating the majority (which I understand is because you feel that the Herald has not projected overhang as ACT has achieved 0.45%)
• The majority required in a parliament in 120 is indeed 61.
• 61 seats is indeed achievable with by the sum of the median seat projections of National, NZ First and ACT alone, which I do understand.
• However, by the same logic, why is it unacceptable to create a majority with the sum of the median seat projections of Labour and Green and Maori which also comes to 62, which clearly exceeds 61???

-Sleepingstar (talk) 09:13, 16 September 2017 (UTC)

@Sleepingstar: - All I did was reformat the footnote. The editor before me (@Langholz8:) edited the NZ Herald projections. Lcmortensen (mailbox) 10:22, 16 September 2017 (UTC)

@Lcmortensen: My most sincere apologies, this was an honest mistake, the previous message was directed to User:Langholz8. Sleepingstar (talk) 10:36, 16 September 2017 (UTC)

Assumptions behind NZ Herald projection calculations

@Langholz8: I've just been carefully reading through the editing summaries you have provided thus far, and having a good long think about them. I would also like to invite @Elcalebo:, @Lcmortensen:, and all other active contributors to the template to weigh in as well.

Langholz8 Edit Summary 1:

“ 01:26, 16 September 2017: Herald do not predict overhang sear ”

Response: I agree, 0.45% entitles ACT to have 1 seat, thus there is no overhang according to the Herald's individual party vote shares. (However, as we have also stated in the notes section repeatedly, the medians do not add up to the total number of seats, because they use independent modelling for the vote share for each party, instead of using a single consistent and congruent model for all parties).

Langholz8 Edit Summary 2:

“ 05:47, 16 September 2017: ACP has 0,45 % - no overhang seat - straight forward calculation show National - ACT-Maori -NZ have 61 seats ”

Response: As above, yes 0.45% entitles ACT to 1 seat, thus there should be no overall overhang. However, National (51) + ACT (1) + Maori (1) + NZ first (9) is actually on 62 seats. While National (51) + ACT (1) + NZ first (9) is on 61 seats, as you have commented.

But there appears to be a mistake on the column as it stands now which states that National + ACT + Maori + NZ first is on 61 seats - I have thus fixed this to state they are on 62 seats.

Langholz8 Edit Summary 3:

“ 13:30, 15 September 2017: Delete impossible majorities ( Labour - Green majority implies that National-NZ-Maori -ACT are minority - average seat paradox !! ”

Response: The so-called paradox isn't really a true paradox. Rather, it is simply based on two different sets of assumptions. The first being that we simply use the vote shares from the individual parties of the NZ Herald projection, and as stated above, the party vote shares would indicate that parliament should not have an overall overhang, thus the total number of seats is 120 seats.

However, it is very clear that in calculating the overall results cell, all editors, including yourself and myself, have been using the median seat projections for each individual party. However, as a result of using median seat projection of parties that are independent modelling, the median seats do not actually add up to 120 seats, instead they actually add up to 123 seats.

Therefore using median seats, the 61 seats achieved by Labour (55) and Green (6) is actually not a majority if we are using the median seat projections which add up to 123 seats. It would only be a majority if one assumes parliament is 120 seats for the purposes of this calculation. But as you have pointed out this is unsuitable - given that this would preclude an alternative governing majority.

This suggests that it would not be congruent to use the median seats in the overall results calculation and ALSO make the assumption that the total number of seats is 120. The two sets of assumptions are actually incompatible in the current situation.

The question now is therefore which assumption should over-ride the other.

Firstly, the Herald explicitly states the median seats of each party from their individual modelling. However, the Herald does not actually explicitly state the total number of seats of parliament. This was an implicit corollary that we have derived by looking at the individual party vote shares.

Secondly, it appears all editors, yourself and myself included, have been using the median numbers of each individual party. We haven't actually adjusted the median seats to force them to add up to 120.

Both points would suggest that the explicitly stated median seat projection calculations should most probably over-ride the implicit corollary that the total parliament seats is 120 and 61 is required for a majority.

As a result 61 should ideally not be used as the majority threshold, as it is, as it stands currently, is incompatible with the median seat projections, because it results in inaccurate results (such as the "impossible majority" of Labour-Green as you have already pointed out). Instead one should probably take into account that the median seat adds up to 123, and use 62 as the majority threshold, in order to be on the safe side. This is not the ideal situation (because we are creating an overhang when there shouldn't be one), but is necessary if we choose to use the median seat numbers that the Herald have already explicitly provided for us.

An alternative (but probably even less ideal) solution is to adjust seats to force the total to add to 120. In this situation we are choosing to use the implicit corollary that there is no overhang in parliament in preference to the median numbers explicitly reported by the Herald. One could potentially punch the Herald vote shares into the official New Zealand Electoral Commission's MMP calculator ([1]). Personally, I feel that this is probably much too onerous, and potentially very confusing for everyone since the these numbers will be different from those explicitly reported by the Herald. I suspect that Langholz8 may have potentially done these calculations, but I feel that if you are to use them, then you must adjust the numbers in the table accordingly, to make the exercise transparent. Furthermore, I'm not currently aware of any clear cut method of transforming the upper or lower seat limits for each party if one was to adopt the exercise of transforming all projections such that they fit into 120 seats, should we adopt this alternative proposal.

Langholz8 Edit Summary 4:

“ 01:14, 16 September 2017: Labour -Green- Maori on have 60 Seats ”

Response: Sorry, but this is unfortunately simply not true, if one is using the median seat projections provided by the Herald. Labour (55), Green (6), Maori (1), add up to 62 seats. 62 seats would fulfill a parliamentary regardless of whether we use a total of 123 seats (derived from the median seats in parliament reported by the Herald), or 120 seats (derived from the implicit corollary that there is no overhang).

If, based on the Herald's explicit median seat projections there is three scenarios of producing a government majority: (1) Labour-NZ First (64), (2) Labour-Green-Maori (62), (3) National-NZ First-Maori (62), then it is important to reflect this in the overall results cell, because there are huge political implications and ramifications.

In light of all of the discussion above, and all the counter-arguments I am to generate in my head at the moment, I have decided on editing the herald column to show all three scenarios of potential government majorities, and using 62 as the majority threshold, and sticking strictly to the median seat projections reported explicitly by the Herald, in favour of the implicit corollary that there is no parliament overhang. I acknowledge and recognise that this situation is less than ideal, but I'm afraid this is the best that we can come up with at themoment.

If you disagree with the above, please, please, please be kind enough to reply to the talk page first if you find yourself tempted to hit the revert button, so we can fully understand your perspective.

-- Sleepingstar (talk) 21:00, 16 September 2017 (UTC)

The point is quiet simply; Maori have 1 seat - therefore they need 59 of the remaining seats 119 for majority- 59/119 >0,5 --> Maori can form only government with the biggest bloc : National+NZ+ACT or Labour+Greens. Your calculations do not work. A Labour-Green-Maori- majority assumes that we are in one of the simulations with greens in parliament- therefore standard median seat can not be used. Best thing to do is calculate seats based on mean vote share to decide majority. --188.22.254.28 (talk) 22:40, 16 September 2017 (UTC)

Alternative proposal for Herald Column

@Langholz8: (Are you IP 188.22.254.28?) From what I understand, your approach appears to be using the mean vote share to back-calculate the central tendency of seat allocations, which was the second alternative solution proposed above. While understandable, it still rests on the arbitrary assumption of where to set the value of central tendency for each party.

I propose a third alternative solution, which I also invite all active contributors (@Elcalebo:, @Lcmortensen:) to the template to weigh in on as well.

Given that the Herald lists a range of possible situations/ simulation results, the most accurate method of calculating the overall result using values explicitly given by the Herald (and also the method making as few assumptions as possible), would be to list the range of possible majority arrangements using all possible simulation results, of which again there are three scenarios.

In essence, I feel like we need to stop fixating on trying to report the value of central tendency and instead the report the full range of possible outcomes from the Herald projections:

• National+NZ First: combined the two parties would fall have 60±5 seats, hence could form a majority between 61-65 in the upper limit of their projection.
  • This is possible because in those circumstances the other parties would all fall within their projected range - Labour on 53-57 (projected 55±3), Green would be on 0 (projected 0 or 6), Maori 1, ACT 1.
• Labour+ Green: Labour is projected to be on 55±3 and Green projected to be 0 or 6. So they can achieve a majority between 61-63.
  • This is possible given that National and NZ first will be on 55-57 (projected 60±5), Maori 1, ACT 1. All aprties fall within their projected range.
  • Note that Labour-Green cannot get to 64 (58+6) seats, as this would mean that national + NZ first will fall to 54 seats, Maori 1, ACT 1. This is impossible as National + NZ first on 54 seats would fall outside of the projected range of 60±5 for the two parties.
• Labour + NZ First could muster up to 64 ±6, so they could form a majority of 61-63 or 65-69
  • This is possible as in this case National would be fall between 49 to 53 (within projected range 51±2), Green 0 (projected 0 or 6), ACT 1, Maori 1.
  • OR National 49-51 (within projected range 51±2), Green 6 (projected 0 or 6), ACT 1, Maori 1.
  • Note that Labour + NZ First can not achieve 64 seats, as in this situation National would not fall within its projected range of 51±2 regardless of whether Green is on 0 or 6, as National would be on 54 or 48 seats - outside of the projected range of 51±2.

Therefore, overall, using the least number of parties to create governing majorities there are three scenarios, with range listed.

1. National+NZ First: 61-65 seats.
2. Labour+NZ First: 61-69 seats. (Albiet, more accurately, 64 seats is impossible)
3. Labour+Green: 61-63 seats.

I have therefore made the proposed changes to the Herald column. Please do reply again if you find the changes unacceptable, but I feel like we have made quite a bit of progress (it was definitely quite helpful when you clarified your thought process a bit more). --Sleepingstar (talk) 23:44, 16 September 2017 (UTC)

They are unacceptable - obviously you again ignore simple math--188.22.254.28 (talk) 23:58, 16 September 2017 (UTC)

Please read the proposal carefully, I feel that you haven't yet fully grasped the rationale behind them. It shifts away from trying to calculate a value of central tendency (which doesn't work because the Greens have an assymetrical distribution of only 0 or 6 seats), and instead focuses on looking at the full range of possible simulation outcomes from the Herald piece. --Sleepingstar (talk) 00:03, 17 September 2017 (UTC)

You ignore basic probability theory. unacceptable --Langholz8 (talk) 00:05, 17 September 2017 (UTC)

Please clarify that in detail please. I don't think the last two remarks made were particularly conducive towards achieving understanding behind your rationale or perspective. The biggest issue I have with your above stated approach is that the Greens being projected to have a non-parametric distribution means that simple addition of medians or means don't work. Sleepingstar (talk) 00:10, 17 September 2017 (UTC)

When you want to predict a Labbour-Green-Maori Majority you need to estimate the number of seats given that fact that greens enter parliament. Obviously in this case the other parties (including Labour) have less seats--Langholz8 (talk) 00:16, 17 September 2017 (UTC)

@Langholz8:As stated above the situation where Greens are on 6 seats, then if Labour holds onto 55-57 seats (projected range 55±3), and if National and NZ first falls to 55-57 (projected 60±5), and Maori 1, ACT 1, then they would meet a majority, and all parties add up to 120.

Again, we note that it is impossible for Greens to be on 6 seats, Labour 58 seats, as this would mean with Maori 1, ACT 1, National and NZ first would be on 54 seats (oustide of the porjected 60±5).

So yes, according to the ranges provided by the herald, labour and Green can form majorities when they are closer to their upper limits, and NZ first and NAtional are closer to their lower limits.

Furthermore, it is possible for National-NZ First, and labour-NZ first to hit the majority of 61 seats, when they are closer to their upper limits, and the other parties closer to their lower limits. --Sleepingstar (talk) 00:20, 17 September 2017 (UTC)

The most reasonable way to calculate majority is to use median percentage - Why do you want to change that ? - If you want to add margin of error - you need to do it for all polls and not just for NZ Herald --Langholz8 (talk) 00:23, 17 September 2017 (UTC)--Langholz8 (talk) 00:23, 17 September 2017 (UTC)

@Langholz8: The NZ Herald is the only source that states a range for predicted seats, all other sources provide a single value for each party. It looks more like a stated range of possible seat projections, rather than simply being an margin of error/ SD/ 95% CI.

I know there is a temptation to use the central percentages. However the values you get from the calculations will be different to the medians stated in the Herald, which is confusing. I'm not sure whether using the median percentage is the most "reasonable", because everything is tehcnically probabistic distributions, and this method creates issues with the Green's because the seats will become non-parametrically distributed because of the 5% threshold. --Sleepingstar (talk) 00:25, 17 September 2017 (UTC)

When we do not agree maybe we should leave the spot empty.--Langholz8 (talk) 00:47, 17 September 2017 (UTC)

@Langholz8:Sure, why don't we blank it out for now, and see what the others (@Elcalebo:, @Lcmortensen:) think. Sleepingstar (talk) 00:49, 17 September 2017 (UTC)

Most likely the next polls increase green vote share and the discussion is obsolete--Langholz8 (talk) 00:56, 17 September 2017 (UTC)

@Langholz8: One can only hope so. The assymetrical non-parametric projection of the Greens is a complete nightmare to deal with. -- Sleepingstar (talk) 01:03, 17 September 2017 (UTC)

@Sleepingstar: I agree with your suggestion - I say go for it! It is a bit of a nightmare reporting this, especially the way the medians don't add up to the total number of seats, and the Greens' asymmetrical estimate. It doesn't help that they don't seem to be reporting the Greens' upper estimate, which should be about 7, given that their upper estimate is 6% of the party vote. Thanks for your work and idea. Elcalebo (talk) 04:41, 18 September 2017 (UTC)

Alternative Solution

I don't think the current way is best. I think the best solution is : Majority Labour +NZ and add remark that it is is uncertain - either Labour-Green Maori have 61 seats or National ACT NZ have 61 seats--Langholz8 (talk) 00:36, 17 September 2017 (UTC)

@Langholz8: I am open to this option, but only on the condition that we try to make it consistent for all the other projections, including the Radio NZ and Horizon projectoins. I'm assuming that labour + NZ First is chosne because they have the largest majority of the three scenarios? In that case, with Horizon and Radio NZ, should probably only highlight the combination with the greatest majority, and then foot note the smaller majority combinations... -- Sleepingstar (talk) 01:02, 17 September 2017 (UTC)

I dont understand the point in the Horizon case - it is a poll - not a poll average - and they have 62 seats even without Maori --Langholz8 (talk) 01:25, 17 September 2017 (UTC)

I chose Labour+NZ because it is the only undisputed majority. - The Maori coalitions are highly unclear - in particular the Maori coalitions highly depend on the Greens.--Langholz8 (talk) 01:41, 17 September 2017 (UTC)

@Langholz8: The horizon seat projections are based on a single poll, just like 1 news, Newshub and roy morgan. It's right at the bottom of the page [2]. --Sleepingstar (talk) 01:59, 17 September 2017 (UTC)