Bias can be a terrible thing. In the past it led to to the Literary Digest predicting Alf Landon would beat Roosevelt in the 1936 U.S. presidential election, Ford thinking the Edsel was a winner, and to Marty Cooper (the inventor of the mobile phone) to say that “cellular phones will absolutely not replace local wire systems.”
However, there are a number of myths about bias, which are themselves a form of bias, and which need addressing. An understanding of these biases is critical for companies that use customer intelligence tools and data to make better business decisions.
In this post, I describe five key myths about biases that you need to be aware of in research and insight.
Myth 1: Some research does not suffer from bias.
All research suffers from some form of bias. Some people think that scales can be taken from academic research combined with random probability sampling and the result will be “objective” or bias-free research.
In reality, however, bias enters the research process at every step:
- When a topic is picked for research, other possible topics are not picked. That is a subjective decision, i.e. bias.
- When the research is designed, a host of biases enter the picture. Social desirability bias means questions like “How often do you clean your teeth?” attract overclaim, in that people tell to "round up" their answers. Acquiescence bias is when people tend to agree due to the way the question is asked. Order bias means that people are more likely to pick the first item on a list. Framing effects mean that the competitive set we ask about changes the answers.
- The people who take part in the survey are a source of bias. The people we contact are one bias. Whether they are willing to be interviewed is a bias. Whether they have been interviewed before (by us or by somebody else) is another bias.
- When the data is analyzed and the insight is extracted, there are several subjective elements, all introducing bias. Not all findings can be reported, a coherent story needs to be created—this is subjective. The decision about what to leave out and what to leave in are biases. For example, confirmation bias means we are more likely to accept data that agrees with our initial idea than data that disagrees.
A great example of the problem is the research disaster that was the U.K.’s 2015 general election. The polls all got it wrong—badly wrong. There were essentially two types of polls: polls using Internet access panels and those where people were phoned using random digit dialing. Both systems were equally wrong. People who were willing to answer polling questions differed (it turns out) from people who were not willing to answer the questions.
The bottom line: There is no gold standard. Every research method has problems and these need to be assessed. Trade-offs need to be made between speed, cost, relative accuracy and type of bias.
Myth 2: All bias is bad.
Well, if we call it bias, it certainly sounds bad, but that language hides a host of key benefits that are associated with what we call bias. Consider the following:
- When measuring customer satisfaction, researchers normally restrict their sample to customers. Indeed, CSAT measurement is normally restricted to customers who have recently used the company’s service or product; this is bias. The alternative is to interview non-customers and customers who have not used the services recently; however these people can’t provide informed information and this adds noise to the signal.
- When running a focus group or online discussion, researchers want people who are willing to take part in discussions—people who will actually share views. A “fair” sample of the public would result in many people who would tend not to join in the discussions. So, researchers compromise and thereby get a richer conversation.
- When choosing researchers for a project, companies normally choose professionals who know already something about the subject. This can lead to confirmation bias, but the alternative of using people who do not already know something about the subject and the techniques used would add noise and be less likely to work.
- Projects that require effort, such as maintaining a mobile/video diary, means that researchers exclude the people who can’t be bothered to participate. This bias is unavoidable because researchers can’t make people do our research.
- Creating an online discussion or community tends to attract people who want to be heard. They may be angry with the brand, love the company, think they have a great idea, want to get something off their chest or have any of a thousand reasons. However, people who don’t have a view or are not likely to change their behavior are less likely to join. Again, focusing on the people who have a view and who may act on that view focused this the research on people who are commercially more relevant.
A purist approach to research would say “ask everybody” and “ask everything,” and then filter out the noise. But this approach is costly and too slow. More importantly, the process itself doesn’t get rid of bias completely. In market research, bias can be a way of increasing the signal and reducing the noise.
The bottom line: We can and should use what we already know to increase the signal and decrease the noise, trading-off some risk against a lot of cost and delay.
Myth 3: There is nothing we can do about bias.
The first thing you can do about bias is to recognize that it is all around us. It exists at every stage of every project. The next steps are to evaluate the sources of bias in a project and determine the best way of dealing with them.
A good starting point is asking people from outside your comfort zone to challenge you—to play devil’s advocate. Researchers should also look at past mistakes, and check the published literature on things like question and project design.
"Bias should be acknowledged, recognized and controlled, wherever possible."
Some biases can be addressed directly. For example, questions can be adjusted to reduce bias. The sample can also be checked to see if key groups are missing. For example, you could check if a certain demographic (like those 70 years and older) are often under-represented in research.
With other biases, the best treatment is to hold them constant. For example, you could use the same sample source for each study or use the same questions and apply the same analytics.
Finally, keep the bias in mind during analysis and try to use approaches such as triangulation to support or challenge findings.
The bottom line: Bias should be acknowledged, recognized and controlled, wherever possible.
Myth 4: Insight communities are too biased for “real research.”
Ten years ago, this was a common comment in the research industry. But over the last few years, as the number of organizations using insight communities has rocketed, the concern has shifted from using them at all to whether they can be used for most research purposes. The ESOMAR Answers to Contemporary Research Questions suggests that large, ongoing insight communities are suitable for most types of research projects.
Applying all the points mentioned above, many organizations are using insight communities for almost every type of research. It’s easier to list the types of research they can’t be used for, with the key ones being:
- Market sizing: Estimating what proportion of the population use which products or services.
- Media consumption data: For example, estimating what proportion of the population are watching a particular program or downloading a particular song.
- Research specifically on non-customers: Most communities focus on customers, so if the research needs to focus on non-customers, it is often not a good option.
Two areas where many organizations do not rely solely on their insight communities are brand tracking and customer satisfaction tracking. Organizations that use insight communities for these two areas are typically operating large communities and benchmarking changes in their scores rather than assuming that the absolute values are meaningful.
Many users of insight communities also run a small proportion of their research via other channels (for example via an online access panel) so that they can get a sense of whether there are differences. They then take those differences into account when analyzing the data.
The bottom line: With care, most forms of research can be conducted via an insight community.
Myth 5: Big data will remove the bias in research.
We know that many forms of questions create bias. For example, the question “How many calories did you eat this week?” tends to elicit answers that are distorted by social desirability bias and by lack of knowledge. This fact has led some to feel that observation will provide a bias free answer. However, the process of observing is also a sources of bias, in the choice of what is observed, the impact of being observed, and in terms of interpreting what is happening.
The choice of sample impacts the answers you get. For example, the aforementioned UK polling disaster was partly caused by sampling too few over-70 year olds. This has led some to advocate big data approaches that are closer to being a census, thereby providing a bias-free solution.
Big data will do many interesting things, but it won’t remove bias. The first challenge is that when something is measured, it changes. A great example is what happens to motorists if traffic cameras are erected. When we know our behavior is being monitored, we sometimes change our behavior, resulting to bias.
Big data throws up more spurious correlations. In many cases, the decision about which are spurious and which are meaningful is subjective, based on our prior assumptions. For example, the U.S. has more guns and more homicides per person than most other countries. However, the interpretation of that data tends to depend almost entirely on one’s beliefs (bias) about guns.
The bottom line: As a source of customer intelligence, using big data doesn’t remove bias and could, in fact, result to spurious conclusions.
Bias is unavoidable. In fact, as I discuss above, bias can often result to more effective and efficient market research practices. Being aware of these myths about bias can help you get more accurate insight when working with insight communities and other types of customer intelligence tools.
Thank you for reading my five myths about biases. Do you have any that you’d add to the list?