7 Statistical Terms You’ll Be Glad You Know

The complex mathematical science of statistics can take years of study to master. The vast variety of methods, their proper procedure and application, and accurate interpretation of results can be challenging. Experienced researchers utilize statistics to uncover truths about data, but the expression of statistical analysis can sound cryptic or misleading to those who do not have some knowledge of this math wizardry.

The vernacular used in statistics can be a stumbling block to those who work with these concepts every day, but simply don’t know what they are called. To help bridge the gap between what you probably already know and what statisticians mean by their terminology, here’s a short list of the most commonly-used statistical terms. Understanding these few key terms and ideas can help greatly in knowing the application and use of statistical information and – more importantly – ensure you do not misinterpret a critical report or fail to grasp the importance of some piece of data.

• DESCRIPTIVES vs. ANALYSES: Simply put, descriptive statistics tell you what the data looks like while statistical analyses tell you what the data is like. It’s not dissimilar to meeting someone for the first time: you first take in their outward appearance – physical traits and features, but as you get to know them you learn of what is deeper down – their personality and behavior. So it is with statistics: your earliest exposure is the descriptive measures that help you understand its appearance, and as you dive into the analyses you learn what the data is like on the inside, what variances it contains, and what effects may be at play.
• SAMPLE vs. POPULATION: A sample is the subject group you gather your data from. It is called a sample because it is a representation of a much larger population of interest. For example, let’s say you want to understand how college students choose cell phones. Since you can’t practically study the entire population of interest (every college student in the world) you gather a convenience sample by recruiting local college students. Ideally, you would want a perfectly random representation so that you have a higher likelihood of unbiased results, but as that is virtually impossible, researchers do the best they can to randomize the sample within the study’s confines. Meaningful results can still be acquired from such a sample, but as with much in statistics, you are playing the odds.
• COUNTER-BALANCE: A counter-balanced study is one in which the different conditions have been equally and randomly distributed so that when later statistical comparison is made, the numbers can be meaningfully compared. In usability research, counter-balancing is often used to make sure that the order in which two competing devices are presented to a user does not affect the overall ratings for those devices across users. For example, users may have artificially-boosted ratings for the first device simply because it was a new, shiny concept – something that wears off when they get to the next device. For this reason, you would want to insure that both devices are seen first equally, so that this “new and shiny” factor is evenly distributed across both devices when you look across users. In a perfect world, your carefully controlled study will have no other influencing effects aside from the one you are interested in. In practicality, there is only so much you can do to prevent outside influences, but counter-balancing is one of the means researchers have to help balance those influences so that, at the very least, they are equal between the groups being studied.
• LIKERT-TYPE SCALES: These are a very common measure used in both commercial and academic research. Two things are particularly important to understand about these scales: first, they are a self-report measure, which means they are a subjective measure and have some pros and cons as described further along in this article. Second, they vary widely depending on their application, but generally are always used to glean an understanding from participants about their opinions that can be compared to others answering the same question. A Likert-type scale is typically 3-9 distinct points used to help participants judge reactions or feelings. For example, let’s say you want to know what those college students from earlier think about two different products. After walking them through some task on each product, you have them answer the same question: how easy was it to do the task on this product? They then must select a number between 1 (easiest) and 5 (hardest). Once you have all of the students in your sample go through the tasks and answer this question, you now have data! You can compare how the two products were rated, and run statistical analyses to determine if there was a significant difference between them.
• SUBJECTIVE vs. OBJECTIVE: In short, the difference between subjective and objective measures can be as simple as the point of origin. If the data originates from within the person you are studying, the measure is subjective. If the data originates from without, it is objective. This is an oversimplification, but it’s a start. It is unlikely that you will actually see these terms used in a report, but they are critical in order to understand the value of a given set of data. If you have your participant fill out Likert-type scales and other survey questions, you are gathering subjective data. This data is very helpful for getting inside of the participant’s head, but it’s important to keep in mind that this data is inherently biased, and it is impossible to exclude outside influences from potentially contaminating the results. Objective data, data that is measured from outside of the participant, can be much more bias-free. For example, if you happen to record the speed at which the students perform the tasks on those two different products, or if you note the number of errors each participant committed in each condition, you have also acquired objective data. This data doesn’t include nearly the level of bias that is found within subjective data (though it is not truly possible in our imperfect world to collect data that is 100% free of any bias). Which of these types of data is more or less valuable to you depends entirely on your research purposes. For most applications, a combination of the two will provide you the best picture.
• AVERAGES: Also referred to as the ‘mean’ (but not because it’s malicious), an average is perhaps the most common statistic used in reporting. It is an excellent measure of central tendency, and noting differences in averages can be a good way to eyeball your data. However, it is important to remember that this number does not express statistical significance. The average is a descriptive statistic, which means (see what I did there?) it is not indicative of an effect or the result of statistical analysis, it simply describes the data you have collected. If you really want to know if the differences between averages for your study are significant, you will need to run some statistical analyses on the raw data. Analytical tests, such as ANOVA and t-tests, can uncover relationships between the averages you are comparing and determine whether or not there is a statistically significant effect at play. That being said, a lot of research, especially commercial research, may be more interested in central tendency than other statistical methods. Central tendency literally is the core tendencies of your given population. Such central tendency measures are the shape of the picture you are trying to paint, and they can provide a jumping-off point for further questions and insights.
• STATISTICAL SIGNIFICANCE: This is the ultimate judge in academic research. If your data proves to be statistically significant, the effect you are studying has proven to be true and genuine. Or, at least, the odds that you have discovered a true effect are pretty good. In fact, to pull the curtain back on this mysterious idea, the number provided that describes statistical significance is just a percentage represented in decimal form. For example, p = .05 is the typical benchmark for statistical significance. This number means that there is only a 5% chance that the effect being described has occurred because of random chance. This definition is a bit of a simplification, but it’s a good way to remember it. Statistical significance, or a lack thereof, can be telling, but be aware that “p” can be manipulated by a quite a few variables (see: https://medium.com/humansystemsdata/how-to-make-a-significant-study-734592ab1f10). While the number can be very insightful, I would be wary before betting the farm on such a small piece of the puzzle. Instead, look at the context, figure out what other statistics are saying, and get a larger picture of what is going on with your data before making final judgements.

Statistics may seem to be a vast, complicated sea of math wizardry, and it is that, but it also has a surface upon which the truth lies. By getting down some of the basics, you can glean the truth from a safe, dry distance, and make that report to your boss sound simple while also making you look like a smarty-pants. More importantly, you can ensure that you are not being misled by a fancy-looking number and avoid misunderstanding what the data is actually trying to tell you.

Mikey Brogdon has an MS in Human Systems Engineering and has a particular passion for writing. He enjoys studying how people interact with technology, especially if that technology is in an automobile. He currently works with an expert team of UX professionals at Human Interfaces, Inc. to develop custom research solutions for any UX challenge. If you need help from a full-service UX research consultancy for a study, recruitment, or facility rental, visit our website <www.humaninterfaces.net>, send us an email <https://www.humaninterfaces.net/contact.php>, or connect with us through LinkedIn <https://www.dhirubhai.net/company/human-interfaces-inc>.