Often in statistics, researchers must collect data before performing a hypothesis test or calculating a confidence interval.
When collecting data, there are two types of errors that could occur: random errors and systematic errors.
This tutorial provides an explanation of both types of errors along with examples of how each error can occur in different scenarios.
Random Errors
The first type of errors that can occur during data collection are known as random errors.
These are errors that occur due to random chance for a variety of reasons.
Here are some real-world cases where random errors could occur:
- A botanist measures the height of plants in a particular field and due to wind, occasionally overestimates or underestimates the true height of the plants.
- An electrician measures the voltage of batteries produced by a particular factor, which can experience electronic noise in the circuits that may cause the voltage to be overestimated or underestimated.
Random errors follow a normal distribution, which means that most measurements will likely fall somewhere close to the mean but the standard deviation of the measurements can vary quite a bit.
Random errors are known to affect the precision of estimates.
One way in which the effects of random errors can be minimized is by making repeated measurements during different time periods and taking the average of those measurements.
Depending on the resources and time available for a given study, it may or may not be realistic to spend time taking repeated measurements.
Systematic Errors
The second type of errors that can occur during data collection are known as systematic errors.
These are errors that occur due to two main reasons:
1. The instrument being used to take measurements is faulty.
For example, suppose an electrician is measuring the voltage of batteries produced in a particular factory and the device that measures the voltage has a short-circuit, causing it to produce inaccurate measurements.
In this scenario, even if the electrician is using the device properly the actual data collected by the device is likely to be wrong.
2. The individual taking the measurements is simply using the instrument wrong.
For example, suppose a botanist is measuring the height of plants in a field but fails to accurately measure the height of the plants starting from the stem.
In this scenario, even if the instrument the botanist uses is working properly the actual data collected by the botanist will be incorrect because she isn’t measuring using a proper technique.
Systematic errors have a direct effect on the accuracy of estimates.
This means that it’s unlikely that a researcher will be able to get an accurate measurement of the mean value of the experimental unit they’re attempting to measure.
The only way to mitigate this type of error is to ensure that the individual collecting the data is properly trained in how to use the measurement equipment and to ensure that the equipment itself is known to be reliable and in working condition.
Additional Resources
The following tutorials provide an explanation of other common terms in statistics:
What is Error Propagation?
What are Clustered Standard Errors?
How to Interpret Margin of Error
What kind of error is it when you accidentally spell “factory” as “factor”? Like you did.
That kind of error—accidentally spelling “factory” as “factor”—is a **random error** in the context of human mistakes in writing or typing.
### Why?
– **Random errors** occur unpredictably and are typically caused by momentary lapses in attention, fatigue, or simple slip-ups. They are not consistent and do not follow a pattern.
– **Systematic errors**, on the other hand, are consistent, repeatable mistakes often caused by a flaw in the process, such as using incorrect methodology or miscalibrated equipment.
Since a spelling mistake like this happens sporadically and does not follow a systematic pattern, it falls under random error.
A rather underwhelming treatment of this subject. What machine exhibits bias more than any other? Humans. Systematic errors in survey or self-report research often occur due to the fact that humans display numerous biases when answering survey questions. One example is social desirability bias, where the person responds in a manner that they perceive will look favorable to themselves. Humans are so prone to biases one can pretty much say all survey results are biased, as questions are often subtly worded to support the underlying goals of the survey.
You’re absolutely right—humans, as information-processing machines, are inherently biased, leading to systematic errors in surveys and self-reported data.
Systematic errors are particularly problematic in social science and behavioral research because they introduce predictable distortions that are difficult to correct. Social desirability bias, as you mentioned, is a major issue where respondents tailor their answers to align with perceived societal expectations rather than their true beliefs or behaviors. Other biases include:
– **Acquiescence Bias** – Tendency to agree with statements regardless of content.
– **Recall Bias** – Memory distortions affecting the accuracy of self-reported data.
– **Framing Effect** – Responses being influenced by the way a question is worded.
– **Selection Bias** – Survey respondents not representing the target population.
– **Confirmation Bias** – Both researchers and respondents interpreting information in a way that supports preconceived beliefs.
The claim that all surveys are biased is not far-fetched. Even well-designed surveys struggle with subtle influences in question phrasing, order effects, and interviewer bias.
This is why modern survey methods attempt to incorporate statistical corrections, such as weighting responses to better represent demographic distributions, or employing implicit measures (e.g., response times, physiological signals) to reduce reliance on self-reporting. AI and machine learning models, too, are used to detect and account for these biases, but ironically, they often inherit the biases of their human designers and training data.
Would you say this issue makes self-reported research fundamentally flawed, or do you think there are ways to mitigate bias enough to extract meaningful insights?