My above-the-line Senate preference explorers (NSW, Vic, Qld, WA, SA, Tas, ACT, NT) now have an option to generate cross tables, which should make some systematic investigations much faster, and also makes for easier exploratory work. This post is a short guide to their use.
The new (pseudo-)link appears just above the main preferences table:
|asdf||asdf asdf asdf asdf asdf asdf asdf||asdf asdf asdf asdf asdf asdf||asdf asdf asdf asdf asdf asdf||asdf asdf asdf asdf asdf asdf|
When you click "Generate cross table", the page will loop down the right-most column currently highlighted (i.e., containing a shaded red cell), effectively clicking on each cell in the column. (If no cells are highlighted, it loops down the first column.) The subsequent column generated in the main table is then turned into a row of the cross table.
Here is the first part of the cross table with the settings as shown in the first picture; by default it sorts to ballot paper ordering:
The title of the this table is "1 Row 2 Column", indicating that the number in a cell is the percentage of votes preferencing that column's group at 2, given that the primary vote was for that row's group. The "Base" field is the value that would appear in the highlighted cell when clicking down the column of the main table.
Reading along the top row of the cross table, the base field tells us that 2.26% of ATL votes went to Family First. Of those, 33.51% went to Labor (partial donkey votes), 7.48% went to the Greens, 13.26% to the Christian Democrats, and so on. We can check that the main table agrees with these numbers by clicking on the FF (sorting by ballot order):
The cross table can be sorted by row or column by clicking on the relevant header (the group name), and the pseudo-links above it allow you to either change the sorting of the rows to match the column order (and vice versa), or reset to ballot paper ordering. Here is the top-left part of the cross table after sorting by Base, then sorting columns to match row order:
To find out which parties' (ATL) voters preferenced Labor most often, click on the ALP column header:
Greens voters were the most likely to vote 2 ALP (37.2%), followed by Family First (33.5%, those donkeys again), then JLN (20.4%).
Alternatively, we could sort by the JLN row. This will make the JLN row match what we could find from the regular table, but now we can quickly compare the difference between JLN voters' preferences and those of other parties:
The most notable difference is that Lambie's voters preferenced One Nation at much higher rates than other major/major-ish parties – 21.6% percent of 1 JLN votes being 2 PHON, versus less than 7% for each of Labor and Liberal, and just 1% for the Greens.
We're not limited to just looking at the first two preferences. By choosing the table type "Later preferences", we can see the percentage of each party's voters that preferenced each other party in their first 6:
Reading across the top row, it says that 57.47% of Family First voters preferenced Labor in their first 6, 35.75% of them preferenced the Greens in their first 6, and so on.
William Bowe at Poll Bludger has an equivalent spreadsheet (which also has a sheet for below-the-line votes), and there are some small differences between our figures that I haven't reconciled – perhaps some exhausting votes not being included in the spreadsheet:
Sorting the above cross table by column shows which parties' voters preferenced that column's party in the first 6 most often. An interesting case is in Queensland, where One Nation received a much stronger preference flow than expected. For this screenshot I've sorted by the PHON column, then scrolled the table to the right so that PHON appears as the first visible column in the table to the right of the row names (freeze panes!):
It's not surprise to see the Australian Liberty Alliance at the top, with 72% of their ATL voters preferencing the like-minded One Nation in their first 6. Fully 60% of Katter's Australian Party voters did the same, 39% of Lazarus voters. The majors are near the bottom of that screenshot, about 23% from each. Reversing the sort, it's the Greens ATL voters who preferenced One Nation in the first 6 the least (somewhat to my surprise, since there are plenty of left-wing micros whose voters I'd expect to loathe Hanson):
One of the topics that fascinates me most in the Senate preferencing data is seeing all the widely varied political preferences held by people. Of course some just follow a how-to-vote card, and some might be picking parties they've heard of in more-or-less random order, but you can isolate some distinct groups of voters.
Here I will look at Greens voters in Tasmania. I want to compare subsequent preferences for different choices of preference 2. That is, I want to see if 1 Grn 2 ALP voters differ systematically from 1 Grn 2 REP voters, or 1 Grn 2 NXT voters, etc. Here is the setup:
I've clicked on two cells, and the cross table will work its way down the right-most column that I've clicked in (here, "Pref 2"), outputting the next column (the third column in this example, "By 6") to each row. After sorting by base rate (which is the percentage of 1 Grn votes going to that row's group at 2) and sorting columns to match row order, here is the first part of the cross table:
The Greens HTV card had the Renewable Energy Party at 2, but only 20.5% of their ATL voters followed the card even that far. Of those who did vote 1 Grn 2 REP, 80% put Labor somewhere from 3-6, and (not shown in the clipped screenshot, because I didn't sort by the REP row), 75% preferenced the Arts Party by 6, 74% for the Science Party, and 57% for the Recreational Fishers (evidently not so popular even among Greens voters who pay attention to the HTV card).
Systematic differences in the Greens voters' preferences can be seen by looking for substantially different numbers in each column. For instance, only 50% of the 1 Grn 2 Sex/HEMP votes put Labor in the first 6, versus 70% for 1 Grn 2 NXT voters. Over 40% of 1 Grn 2 NXT voters put JLN in their first 6, but only 17-18% of 1 Grn 2 Sex/HEMP or 1 Grn 2 AJP voters.
Some of the sample sizes can get small when you drill down too far, but there were about 2000 voters who went 1 Grn 2 NXT, and about a thousand each for 1 Grn 2 S/H and 1 Grn 2 AJP, so the differences in subsequent preferencing are much greater than what would be expected by chance.
A final note on calculation time: the "Step forward" cross tables should get built in a fairly small number of seconds, but the other table types may take a while, especially on mobile; for the NSW "Preference sources" it takes a couple of minutes on my Galaxy S3. (And for those working with the downloaded 1-6 files for the larger states, you might be waiting 10 minutes or so for NSW.)
Have at it!