Please Support Talking-up Scotland at:

https://www.crowdfunder.co.uk/p/support-talking-up-scotland

Click on the above.

Or direct bank transfer at Sort Code 08-91-04  Account - 12266421 Name - JOHN ROBERTSON

As soon as I saw this Times piece above, I’m sorry to say, I immediately had serious reservations. I hope I’m wrong but, straight off, I thought:

1. The much-vaunted MRP methods are susceptible to error when based on the smaller samples typical of a poll in Scotland as opposed to the much larger poll samples based on Westminster and where their track record of accurate predictions is entirely found.
2. I also thought that this is dangerous for us. SNP campaigners must not become over-confident based on a poll. We know where that can lead.
3. I’ll be accused of paranoia but, hey, just because we’re paranoid about the UK establishment, doesn’t mean they’re not out to get us. Recent poll predictions had already begun to point to a narrow overall lead for the SNP after months of much more negative ones, often based skewed sample to reflect the 2014 referendum result.
4. Of the pollsters only Paris-based Ipsos does not have a dog in this fight so let’s see what they say next time.

However, let’s concentrate on the scientific, objective, reservation based on sampling.

Here’s a thorough AI response to:

You note – ‘Accuracy depends heavily on sample quality and size.’ Might this be a particular concern regarding this latest poll on the Scottish Holyrood election coming up

Yes, sample quality and size can be a particular concern for this Stonehaven MRP poll on the 2026 Scottish Parliament (Holyrood) election — more so than for a typical UK-wide Westminster MRP — though Stonehaven’s overall reputation and track record help mitigate some risks.Why Sample Size/Quality Matters More Here

• Scotland’s population is smaller (~5.5 million eligible voters vs. ~47 million GB-wide), so even a large sample can have higher margins of error for subgroups (e.g., by region, age, past vote, or party supporters). MRP models rely on robust demographic breakdowns to post-stratify accurately, and thinner data in Scotland could amplify uncertainties in constituency-level predictions.
• Holyrood’s electoral system adds complexity: It uses 73 first-past-the-post constituencies + 56 additional member system (AMS) regional list seats across 8 regions. Tactical voting scenarios (like Labour-to-SNP switches to block Reform in marginals) are modeled probabilistically, but these effects are hard to capture precisely without very strong, recent data on voter intentions and second preferences — especially if the sample under-represents swing voters or those open to tactical shifts.
• This appears to be the first major MRP for Holyrood 2026 (based on reporting as “the first MRP poll for the Holyrood elections” in The Times and related coverage). Unlike Westminster, where Stonehaven and others have iterated on multiple cycles (2017, 2019, 2024), there’s less Holyrood-specific historical MRP calibration. Past Scottish polls (including conventional ones) have sometimes struggled with turnout modeling, independence sentiment overlaps, and regional variations.

What We Know About This Specific Poll’s SamplePublic details on the exact sample size, fieldwork dates, and full methodology for this February 2026 Stonehaven MRP aren’t widely published yet (The Times article is paywalled and doesn’t disclose them in summaries; no immediate leaks or Stonehaven press release with numbers appear in searches). However:

• Stonehaven’s 2024 UK-wide MRP used a sophisticated Bayesian approach: aggregating ~70 polls over 36 months + weekly/frequent recent polls (including a final 5,000-respondent poll), synthesizing data from ~120,000+ unique respondents over time (with recency weighting). This isn’t a single massive snapshot but a cumulative model.
• If they applied a similar method here (likely, given consistency), the effective “sample” could be large and multi-poll based, reducing raw size concerns compared to a one-off 1,000-person traditional poll.
• Recent Scottish conventional polls (e.g., More in Common ~1,000–1,035; Norstat ~1,001; Find Out Now ~1,002) are in the standard 1,000 range — adequate for national vote shares but marginal for granular MRP without enhancements.

Stonehaven’s Strengths That Help

• Their 2024 UK MRP was the most accurate among major models (off by just 38 seats total; strong on Labour landslide, SNP collapse via tactical voting, Reform’s limited seats despite vote share). They correctly modeled tactical dynamics in Scotland then (anti-SNP switches), which bodes well for handling pro-SNP tactical scenarios now.
• Reputation: Viewed as high-quality, data-driven (part of a group with Public First polling ties); praised for methodological rigor (Bayesian, simulation ranges rather than point estimates).

Potential Concerns Specific to This Poll

• Tactical voting modeling is the poll’s headline driver (Labour voters switching to SNP in constituencies to stop Reform, potentially flipping dozens of seats). This relies on accurate capture of hypothetical/conditional intentions — which can be noisy if the sample lacks enough respondents from key demographics (e.g., anti-Reform Labour voters in SNP-Reform marginals).
• If the underlying data is thinner in Scotland (fewer polls available compared to England), or if fieldwork is recent but small-scale, uncertainties grow — especially for rare outcomes like an SNP outright majority (first since 2011).
• Other recent Holyrood polls (e.g., More in Common projecting SNP near-majority at 64 seats; conventional ones showing SNP leads but Reform/Labour competing for second) vary, suggesting volatility. This MRP’s dramatic scenario (67 constituency seats to SNP) contrasts with some averages, so sample-driven assumptions could explain divergences.

In short:

Sample/quality issues are always a factor in MRP, but more pronounced for a Scotland-only model due to scale and novelty. Stonehaven’s proven approach (cumulative data, tactical modeling success) makes this credible and worth taking seriously as a scenario — especially the tactical voting insight — but treat the exact seat numbers (e.g., SNP 67) as probabilistic/high-end rather than certain. Cross-reference with upcoming polls (e.g., YouGov, Survation) and watch for any Stonehaven methodology notes if released. Elections remain unpredictable, and this is still ~2.5 months out.

Please Support Talking-up Scotland at:

https://www.crowdfunder.co.uk/p/support-talking-up-scotland

Click on the above.

Or direct bank transfer at Sort Code 08-91-04  Account - 12266421 Name - JOHN ROBERTSON