The winning candidate in last week's St Joseph by-election won with just under 45 per cent of the vote, and only five per cent separated the two leading candidates.
Given how close the results were, one might expect that neighbours in similar circumstances considered their options and narrowly decided between two roughly equally preferable candidates.
However, the reality is that the constituency was extremely polarised. In 40 out of the 43 polling divisions (PDs) either the UNC or PNM candidate bested the other by a wide two-to-one margin.
To illustrate the point using the extremes: One PD in Petit Bourg voted 127-7-5 for the PNM, ILP, and UNC respectively, and one polling station in Aranguez voted 329-43-7 for the UNC, ILP and PNM respectively.
Conversely, the PNM won "PNM" PDs 3,885 to 672, and the UNC won "UNC" PDs 5,685 to 1,692. This is a very different picture from the close 45 per cent-39 per cent result overall.
The numbers therefore do support analysts' readings that we have "returned to tribal voting."
Was St Joseph a run-off?
One week ago, while reviewing the local government election, I discussed strategic voting in cases where a run-off election is not available.
The by-election, coming just two weeks after the local government elections, mirrors a hypothetical run-off election and comparing the results of the two elections in identical PDs is a rare case of what we call a "natural experiment" in the social sciences.
If every voter voted, or didn't vote, the same way that they voted in the local government election, the PNM would have won with 46 per cent, followed by 30 per cent for the PP, and 23 per cent for the ILP.
Indeed, whether influenced by the message of the campaigns or by independent consideration, the results of the St Joseph by-election resembled that of a run-off election in which voters coalesced around the two major candidates.
In particular, the gap between the two major candidates closed with many voters who voted for the ILP now voting for the UNC.
We can be certain that the tightening of the race came because about half of the very same electors who voted for ILP candidates in the local government election voted for Mr Alleyne two weeks later by examining the results by PD in a reasonably high turnout election (53 per cent).
Although more people voted for the PNM in the by-election, vis-a-vis the local government elections, the share of votes that the PNM received was almost identical across all 43 PDs between the two elections.
The majority of the swing in the results came in "UNC PDs", and the ILP vote in the "PNM PDs" was largely unaffected.
Strategic voting in St Joseph
The numbers simply don't allow the possibility that half of ILP voters stayed at home last Monday and that the UNC found 50 per cent more voters that did not vote two weeks before.
Voters who had "gone green" changed their minds.
What this means is that:
1. There is evidence of strategic voting, with voters in the Aranguez/Warner Village district casting ballots for the ILP candidate when the PNM had little chance of winning but voting for the UNC when their vote "mattered."
2. If we continue the assumption that the by-election was akin to a second-ballot, then the ILP "split" the vote in San Juan East and Valsayn/St Joseph.
The PNM won in both districts although a majority of the district would have compromised with the UNC and COP candidates respectively as their first or second choice.
I suggested in an article last week that vote-splitting by the ILP did not change the overall results of the election but I singled out San Juan East and Valsayn/St Joseph as two of the few possible exceptions.
Such a change in allegiance between the two elections suggests that voters in these districts voted more strategically in the second election because they had more information about the consequences of their vote.
An examination of the results also show that the ILP share of the vote was roughly equally spread throughout the very polarised constituency.
This suggests, as Sunday Guardian columnist Mark Wilson wrote this week, that the ILP represents the post-tribal, or non-tribal vote, and that vote is not 23 per cent as interim party leader Jack Warner first claimed, but 14 per cent.
Pre-election polling
Finally, Mr Wilson is correct that the pre-election polls failed to predict a close election and all overstated the ILP support.
He wrote that either the polls were an accurate "snapshot" of a race that swung dramatically in the last week or the polls all suffered from poor or inadequate methodology in what was a complex race to measure.
I found it interesting that pre-election polls for the gubernatorial election in Virginia, the day after the St Joseph by-election, had a similar deficiency.
Most had a wide six- to 12-per cent point spread between the Democrats and Republicans, with the actual result only 2.5 per cent points apart.
There was a similar overstatement of support for the third-party candidate, which may have confounded the results.
Given the variety of methodology used locally, it may be indicative that all the polls differed from the final result in roughly the same way.
The poll which was by far the closest to the final result was consistent with the story that half of ILP voters "came home" to the UNC, while the other half, "true" ILP voters, half of whom live in strong PNM areas and the other half from strong UNC areas, returned to vote for the ILP a second time.
This article was written by Nigel Henry of Solution by Simulation, a private firm that uses computer modelling to provide insight into political contexts.