3.7: Lessons Learned About Variables

There are miscellaneous lessons about variables.

Temperature as a variable is a sticking point. Is temperature truly a ratio variable? After all, there is a zero point – 0 degrees Fahrenheit, 0 degrees Celsius. But that zero point is based on a continuum that is human made. Under the Fahrenheit scale, water freezes at 32 degrees; under the Celsius scale, water freezes at 0 degrees. Under the Kelvin Scale, water freezes at 273.15 K. So, there is no established zero point; it is more like there is an established shared state - frozen water, and the different measuring systems have different zero points. So, there doesn’t seem to be an established zero for temperature, so can temperature be considered a ratio variable? The discussion might be the equivalent of splitting hairs. Suffice it to say, temperature is a ratio variable. Even though there is no definitive ‘zero’ point, once you decide which scale you are using, the number can stand on its own without interpretation. Temperature is a ratio variable unless there is a real need to open this discussion.

Returning to the concept of the variable, always remember that when we say variables vary, we really mean there is a change. Change can refer to a change in level, such as low to high amounts or low to high intensity. Or change can be changing from type to type, such as control to treatment or male to female.

It is crucial to know if your variable is considered categorical or continuous. For continuous variables, it is crucial to know if the variable is ordinal, interval or ratio when selecting a statistical analysis. For correlations, it is crucial to know if the variable is interval or ratio versus ordinal. Interval or ratio variables use equal units, and the Pearson correlation is called for. Ordinal variables use unequal units or ranks, and the Spearman correlation is warranted—more on this in the correlation section. However, for regressions, we usually use a regression for all continuous variables, regardless of whether they are ordinal, interval, or ratio. No, that is not good practice, but this issue will be discussed in later regression chapters.

Use your literature review to determine how past studies scaled these variables. How did studies scale the variation for PTSD, bullying, depression, and intelligence? Inventory the variables and how they varied and if those variations helped predict the outcome variables, then catalogue this information as part of your literature review. I would catalog the following.

1. What was the research question? What are the authors trying to predict?
2. What were the constructs under investigation?
3. Do the constructs make sense in terms of how those constructs could predict the outcome?
4. What were the operational definitions? What was the conceptual basis for the operational definitions?
5. Describe how the constructs were defined as variables. Were there several variables per construct?
6. How many independent variables, dependent variables, and covariates?
7. How were the variables scaled? By type or by level. And if by level, they are continuous, and were the demarcations by ordinal, interval, or ratio?
8. If by type, does the number of categories per type align with the theoretical expectations for the variable?
9. If by level, then what do high and low values mean? Do the values mean high or low in amount or high and low in intensity? What are the score ranges for the variables? Is it clear not only what the low and high scores mean but also what the midrange and all other values mean when measuring the construct?

Cataloguing this information could help sort through your variables as you design your own study.

What usually happens is that researchers tend to add variables. Collapsing them, subdividing them, combining them. This practice happens for many reasons. Sometimes, we combine variables when there are not enough participants per category. For example, for the variable race, we tend to have categories in White, Black, Hispanic, and Asian. However, some contexts do not have participants from all these races, such as certain neighborhoods where the census shows a majority of White persons. So, some researchers may collapse the few participants in the Black, Hispanic, and Asian categories into one “minority group” category. Is this good practice? Possibly, if determining who is Black, Hispanic, or Asian has no bearing on the outcome variable. We will discuss this issue more when we discuss the ANOVA analysis.

I bet reading all of this material gave you more than you could ever want to know about variables. Variables are like your ingredients. We work hard to obtain high-quality ingredients to create a high-quality meal or dessert. We work hard to understand our phenomenon of interest by creating high-quality variables to obtain an outcome that furthers our understanding of the phenomenon.