Sir, – Thanks to Philip McGuinness for clarifying the different data sets he used in his analysis of support for Irish unity (Letters, September 9th).
He notes that: “Some of the NILT (Northern Ireland Life and Times) figures quoted by Prof Burke are fascinating: in 2023 only 29.1 per cent supported unity, but 35.3 per cent would vote for reunification ‘tomorrow’.”
Accounting for the difference between 29.1 per cent and 35.3 per cent is fascinating, in various ways.
First, the percentages come from two separate but related questions measuring different dimensions of support for Irish unity: one asking about long-term constitutional preference, and one about voting in a border poll tomorrow, respectively.
Different questions will generally elicit a different distribution of answers because they are measuring different things. Most of the time, researchers are primarily interested in such substantive differences.
Second, some of the difference between our percentages is the result of sampling error, which is a part of all sample survey designs. Taking into account such error, the observed difference between 29.1 per cent and 35.3 per cent could be as low as one-half a percentage point or as high as 12 percentage points. Systematic measurement error is a third consideration.
The NILT question on long-term preference is likely biased, resulting in estimates that regularly inflate support for union and deflate support for unity. It’s no surprise then that the biased measure of long-term preference gives a lower estimate of support for unity (29.1 per cent) than does the relatively unbiased measure of vote in a border poll tomorrow (35.3 per cent). The same pattern holds in other NILT survey years. In sum, both substantive factors and extraneous factors (sampling design and representativeness, question wording, and so on) explain the difference in any set of percentages.
Mr McGuinness says that “it is important not to put all our eggs in one psephological basket.” But it is also important to recognise that the study of one basket can be illuminating, and the study of many baskets problematic. Using numerous data sets with varying research designs might affect our analysis by multiplying the number of extraneous factors that could serve as plausible alternative explanations of the substantive difference in which we’re interested. Disentangling substance from extraneity is especially necessary when the percentage difference is quite small. – Yours, etc,
MIKE BURKE,
Associate Professor Emeritus,
Toronto Metropolitan University,
Toronto,
Canada.