OPINION23 April 2019

Significance testing is insignificant to modern marketing

Innovations Media News Trends UK

Market researchers’ continued use of significance testing hinders innovation and should be consigned to the industry history’s books, writes Jack Miles.

Risk averse assessment metrics innovation_crop

We all use significance testing – the method to show differences in results aren’t caused by chance. It’s a hallmark of quantitative research and gives research users confidence. This is despite significance testing being irrelevant – ironically insignificant – to modern marketing and market research.

Yes, that’s right. The green circles and red squares used by researchers in communication and concept tests are irrelevant. Significance testing is outdated and has origins irrelevant to modern day businesses. Furthermore, the modern marketing landscape promotes innovation, boldness, risk and embraces failure. Significance testing seeks to avoid all of these while feeding our metric obsession.

Outdated and irrelevant origins
Ronald Fisher is credited as being the founding father of significance testing. Fisher’s breakthrough work took place in 1925. Should we be using a 94 year-old process to make modern marketing decisions? Fisher notably also comes from a medical background. The medical profession is cited by Matthew Syed’s book ‘Black Box Thinking’ as being the least embracing of the fail-and-learn culture currently embraced by tech companies. Whose thinking would research rather be aligned with – Google Ventures in 2019 or the Rothamsted Experimental Station (where Fisher founded significance testing) in 1925?

Opposes modern business culture
The rise of fast moving, fail-and-learn tech companies is threatening corporate giants. To defend against this, corporates such as Coca-Cola and GlaxoSmithKline now promote a ‘fail fast’ culture themselves. Within this, there’s simply not the time to have product development halted due to results which aren’t significantly higher than competitors. The culture of red squares and post-debrief procrastination are now replaced with a fail-fast-learn-revise-relaunch approach. Risk-averse research recommendations based on significance testing don’t fit into this culture. As Boots marketing director Helen Normoyle recently said: “The art of great marketing and great insights and research is a combination of really deep customer insights and data with instinct and intuition.” This way of thinking means we need to look beyond boxes and circles when informing marketing decisions.

Supportive of this, Jake Knapp of Google Ventures claims that ROI on research drops after n=5. This is a sample size where significance testing isn’t possible. Although Ronald Fisher would disapprove, this approach to testing has helped Google Ventures become a $2.4bn business.

Builder of false confidence…
Significance testing is designed to give confidence in research results, such as reassuring creatives that their new advert will outperform the existing one. This is dangerous as we all suffer from overconfidence bias. Do we need a dated approach to increase this false confidence? Placed in the context of living in a society prone to black swan events (events which come as a surprise and are inappropriately rationalised with hindsight), falsely installing further confidence about marketing success is dangerous.

…Or killing confidence altogether
When significance testing isn’t giving false confidence, it’s killing it. John Hegarty claimed that data means advertising is no longer engaging with people’s imagination. To stand out in a cluttered world, advertising needs to be bold and distinctive to the point of making its commissioning marketers uncomfortable. Significance testing and its use of red – the indicator of danger – dissuades bold, imaginative creative work through fear of it failing – a fear that is built based on research methodology, not the dynamics of modern advertising.  

Fuelling an (irrelevant) metric obsession
Digitisation has caused marketing to become obsessed with metrics. Dwell time, likes, shares, CTRs, ‘engagement’ and traffic, to name a few. This has kept the term ‘significant difference’ embedded within metric-obsessive marketers’ lexicon. However, there’s only one measure that really matters to marketers – profit.

Does a positive significant difference mean increased profits? No.

Does a negative significant difference mean decreased profits? No.

Therefore, does significant difference as a measure matter? Arguably not.

None of the above should come as surprise. Why? Because innovation is the cornerstone of our industry. It’s the criteria for industry awards and the focus of industry reports. And at the heart of innovation sits risk and failure – the opposite of what significance testing promotes.

Researchers’ continued use of such a traditional method also hinders how we build the ‘market research brand’. Being a more creative and commercially savvy industry is only a good thing. Restricting ourselves to risk-averse recommendations and overly-focusing on the supposedly ‘significant’ differences between vanity metrics hinders this. So let’s think more outside the (red) box, and move significance testing from industry hallmark to industry history.

Jack Miles is senior director at Northstar Research


5 years ago

Hi Jack, Well done for sticking your head above the parapet. I'm sure that you're expecting huge push-back from quanties and I'll not disappoint. If you are advocating an end to large scale quant, that's an interesting debate. If you are recommending focus on consumer understanding for product and ad development rather than post-testing, you'll get wide agreement. But, if we are doing quant research, we need some method to determine the meaning of "better" and "worse" and sig testing is still fit for that purpose.

Like Report

5 years ago

Well done Jack for flagging this issue. It is a matter close to my heart and has been for years. How often in life do we wait for 95% confidence before making a decision? I think business has become more risk averse ironically whilst simultaneously looking to be more agile. To me it comes back to our role as researchers - our job should be to find out stuff and provide advice on the basis of which clients make informed decisions. If you wait for that Number 95 Bus you could be at the bus stop a long time.

Like Report

5 years ago | 1 like

You are right – there is no real line in the sand, the green and red boxes are a distraction from the story, we get hung up on significant differences, even when those differences are tiny. But datasets are not created equal. Significance testing ensures we aren’t driving decisions based on randomness. A simulation of collected data demonstrates how terribly easy this is to achieve. So I must respectfully disagree with your opinion here - rigour has become increasingly relevant to building data driven businesses. We must stand by it.

Like Report

5 years ago

It's the poor carpenter that blames his tools, Jack. It's not that the tools are antiquated and it's not that they stifle creativity. The modern marketing landscape might embrace failure (I don't agree with you there either) but if true, they'll get there faster when ignoring research. The problem is that research companies often don't know whether what they measure matters - and at the end of the day profit is probably what matters (maybe sales, from a research perspective, is sufficient). That a marketer pays attention to bad metrics or insists on insanely high confidence levels is not the researcher's fault per se, although a good researcher would say, "don't use this tool" if they have reason to believe it will lead the marketer astray. 

Like Report

5 years ago

I couldn't quite understand why you are so against a well established statistical method. Significance tests are useful to identify patterns and find out more about behaviours. Why are the tests needed to be related to the profit? Why do market researchers need to tell their clients only what is profitable and what is not? I don't think modern marketing is only about profit is it?

Like Report