When you weigh yourself in the morning, do you feel that your scale is biased and always weighs you more than you think you should weigh? When timing how long it takes to complete an invoice, do you believe the time is always longer than it should be? 

There are many different types of bias when collecting and analyzing your data. A consistent difference between the observed measurement and expected measurement is called bias. Something is always lower or higher than you expect.

This article will present the different types of bias you might experience, the downsides of having bias in your measurements, and how you might eliminate or reduce your statistical and measurement bias.

Overview: What is bias? 

Bias is the systematic, unidirectional tendency of a measurement to under or overestimate the true underlying value of what you are measuring. 

For example, the scale you use to weigh incoming 50-pound bags of powdered raw material consistently shows a weight of 49.25 pounds. The difference between the measured weight and true weight is called bias. In this case, the bias is 0.75 pounds. 

Or, your preference survey always shows females to be more satisfied than males. Is this true, or did you not include enough males to get an unbiased sample?

The bias of your data and measurement system can come from many sources. Here are a few of the most common ones.

1. Selection bias: The process of selecting data, people or groups resulting in a sample not representative of the population. Included in selection bias are:

• Sampling bias: Non-random sampling can result in sampling bias. You take your samples at the beginning of the shift, but not throughout the day.
• Time interval bias: Sampling from a fixed time period may cause bias. Only sampling on Friday afternoon may result in a bias based on the time you are sampling.
• Confirmation bias: You are more inclined to accept results that confirm your own beliefs.
• Self inclusion: You volunteer to be part of the survey.
• Exclusion: You hang up the phone when you are asked to participate in a survey for a subject you are not interested in. Or, you decide not to collect data from the third shift but base your decision only on data from the first shift.

1. Survivorship bias: You wonder why your fellow gym members seem to be so motivated to attend each day when you don’t feel the same way. Have you also collected data from those who don’t attend each day? They may feel like you.
2. Omitted variable bias: Results from the absence of relevant variables. That beautiful condo on the ocean looks like a great bargain at only $450,000. But, are you aware there is an upcoming $80,000 special assessment to repair rusted beams and columns along with rusted rebar in the balconies?
3. Recall bias: Recall and recency bias occur when you don’t recall sufficient information from the past and place more importance on the most recent information.
4. Observer bias: Scoring gymnasts in the Olympics might reveal judges scoring fellow citizens differently than those from other countries.
5. Funding bias: When a pharmaceutical company sponsors a study for one of their drugs, the results might show some bias since the researchers are getting paid by the company.
6. Non-response bias: Do people answering your survey feel the same as those who didn’t respond? Non-response bias needs to be addressed any time you do a survey.

Is bias always bad? It depends. While bias is not desirable, in most cases, it’s not a disaster. If your scales are underweighing things, you simply adjust the scale. Other ways to mitigate or eliminate bias are:

1. Calibration of equipment 
2. Use of random sampling or stratified random sampling to select objects or people for measurement.
3. Testing of non-response bias for surveys
4. Blind rating or measurement
5. Use of Measurement System Analysis to assess your measurement system
6. Unbiased statistic — when doing statistical analysis, use an unbiased statistic or estimator to represent the population parameter

3 drawbacks of bias

A bias, by definition, is either an over- or under-estimation of the true value or characteristic of what you are measuring. 

1. Incorrect estimation 

The issue is how big of a bias you have. If it’s small, even the biased measurement might be acceptable. Unfortunately, you might not know the degree of bias if all you have to rely on are other biased measurements.

2. Reduced credibility 

If there is a known bias, the validity and credibility of your conclusions might detract from the quality of your decision.  

3. Negative impact based on a wrong decision 

If you make a decision based on inaccurate sample estimations of the true underlying population values, you might negatively impact the organization.

Why is bias important to understand? 

It is important to understand the possible sources or causes of bias so you can eliminate or reduce the possibility of them occurring. 

Bias is systematic or consistent

Bias in a measurement is not a fleeting event but an ongoing, consistent error in estimation. 

Can be mitigated 

Bias is not the cause, but the effect. Eliminating the source or cause of the bias will reduce or eliminate the bias itself. 

Unbiased statistic and estimator

You use sample statistics to estimate population parameters. Be sure you understand the importance of using an unbiased statistic when doing your statistical calculations.

An industry example of bias 

The marketing department of a well-known food company decided to send out an email survey regarding a new product under development. Based on the positive results, they made a decision to bring the product to market. After four months of poor sales, they decided to remove the product.

In retrospect, the VP of marketing realized they made the decision to release the product based on biased and misleading results from the survey. Here was where they felt they went wrong:

1. They didn’t use random sampling to get a more complete perspective from their potential customers.
2. The respondents were self-selected and self-excluded.
3. There was no follow up with the non-respondents to test the non-response bias.
4. They assumed their entire population of possible customers had access to email. 

3 best practices when thinking about bias 

Bias is preventable. Use these best practices to understand how. 

1. Measurement System Analysis 

Conduct a Measurement System Analysis, or MSA, to understand whether your measurement system has a tendency to exhibit bias. 

2. Calculate non-response bias for all surveys 

Those who answer surveys and those who don’t usually have different perspectives. To get a balanced and representative view, be sure to conduct a non-response bias analysis if your response rate is low.

3. People as well as equipment 

People can also demonstrate bias when making subjective assessments or measurements. They will also need to be calibrated as appropriate.

Frequently Asked Questions (FAQ) about bias

1. What are common types of bias for surveys? 

Sampling bias, non-response bias, response bias, and order bias are a few of the more common types of bias encountered when surveying.

2. What are two common biased sampling techniques?

Two of the most common biased sampling techniques you’ll want to avoid are convenience sampling, where you select whatever is close and available, and voluntary sampling, where the subject decides to opt in and is not randomly selected. 

3. What is the best way to prevent sampling bias? 

Use random or stratified random sampling.

Final thoughts on bias 

Bias exists in many aspects of a business. While social bias is important to address, our focus has been on statistical and measurement bias. By carefully selecting your sample of people or items, you can reduce or eliminate the systematic and consistent estimation errors you would get from a biased data set.