In reference to Philip McGuinness’s data analysis of support for Irish unity, I appreciate his clear explanation of the varied data sets he employed (Letters, September 9th). Fascinatingly, the numbers cited from the Northern Ireland Life and Times (NILT) by Professor Burke revealed that in 2023, a mere 29.1% were in favour of unity, whilst 35.3% expressed readiness to vote for reunification ‘tomorrow’.
There’s indeed an interesting disparity between the 29.1% and 35.3% figures; let’s break it down.
Firstly, the figures represent results from two distinct but related questions. Each evaluates a diverse aspect of the support for Irish Unity: one looks at long-term constitutional preference and the other gauges immediate support via a border poll. It stands to reason that different questions would evoke a diverse range of responses since they evoke different aspects; researchers are usually keen to study these voluminous distinctions.
Next, the deviation in the percentage rates partly steams from sampling error, a common feature in all sample surveys. Once this error is factored in, the actual difference between 29.1% and 35.3% could potentially range from half a percentage point to 12 percentage points. Measurement error is another factor to consider.
Specifically, the question on long-term NILT preference is likely skewed, habitually amplifying support for union and diminishing support for unity. Consequently, it makes sense that a biased question about long-term preference would report lower unity support (29.1%) compared to the more unbiased question about an immediate border poll vote (35.3%). This trend is also evident across other NILT surveys.
In conclusion, the percentage disparity can be attributed to both substantive elements as well as external factors like sampling design, question semantics, representativeness etc.
Mr McGuinness expresses his belief that it’s not wise to fully commit to one electoral modelling technique exclusively. However, he also emphasises that while studying one method can provide valuable insights, it becomes more complex when numerous methods are involved. Using different data pools with varying approaches can interfere with our analyses by introducing a multitude of unrelated factors that might present credible alternative accounts for the matter we’re focused on. Particularly when dealing with minuscule percentage differences, it is imperative to segregate actual substance from irrelevant aspects. – Yours faithfully,
MIKE BURKE,
Retired Associate Professor,
Toronto Metropolitan University,
Toronto,
Canada.