Statistics and data presentation: understanding variables
All science is about understanding variability in different characteristics, and most characteristics vary, hence we call the characteristics that we are studying ‘variables. When we work in a quantitative area, we make measurements. The scale of measurement is very important because one criterion for selecting the appropriate statistical technique is the scale of measurement used to measure whatever it is, we are studying.
There are different statistical techniques to use with each kind of measurement.
✓ Nominal Scale is the lowest level of measurement. Sometimes this is referred to as qualitative data – not to be confused with qualitative research. This scale uses numbers to describe names of discrete categories. One determines for each case whether they have or do not have the attribute in question.
✓ Ordinal Scale is used to rank people in order (e.g. least politically active to most politically active). This is the lowest level of quantitative data and involves the process of assignment of numbers to cases in terms of how much of the attribute is possessed by each subject.
✓ Continuous data can assume different values within a range. Interval Scale is where a number assigned is the amount of attribute possessed. Most statistics procedures can be used with interval data. Ratio Scale is considered the highest level of measurement, because all statistics tools can be used on ratio data.
When you read an article, you need to figure out what all the variables are in a study. Then you need to identify three things for each variable one at a time: the scale of measurement; the possible score range; and the meaning of high score and low score. Variables take on different functions in a study. We have to be able to tease these functions out. When you are conducting research, you have to recognize the different variables that are at play in your study so you can account for them during your analyses. Variables can take on different functions within the same study, so don’t classify them at the start. Researchers decide on a classification of variables in each analysis. Let’s take a look at the different classifications of variables.
Classification of variables
• Dependent Variable: The outcome variable of interest is observed to see whether it is influenced by a manipulated variable. This is called a dependent variable. In other words, a characteristic that is dependent on, or thought to be influenced by, an independent variable. This is sometimes called outcome or response variable.
• Independent Variable: In experimental research, the researcher can manipulate one variable and measure the effect of that manipulation on another variable. The variable that is manipulated is called an independent variable. In other words, a characteristic that affects, or is thought to influence an outcome or dependent variable, or an antecedent condition. Independent variables are sometimes called factors, treatments, predictors, or manipulated variables.
In a better scenario, the only consistent feature that varies between an intervention and control group would be the outcome variable of interest. However, this is not generally the case, and we often have confounding or extraneous variables that play a part. When we design our research studies, we need to pay attention to and account for these variables also.
• Control Variable: any variable that is held constant in a research study by observing only one of the instances or levels. Control variables are not necessarily of central interest, but things that a researcher cannot change or remove from participants. They might be known to exert some influence on the dependent variable. We can’t study everything, so a researcher may be interested, for example, in how parental education (and some other variable) is related to reading ability in younger children. He/she happens to know through previous research that gender is related to reading. So, for the purposes of the study, they chose to study only girls. Thus, gender is the control variable and is “held constant”.
• Mediator (Intervening) Variable: a hypothetical variable that explains the relationship but is not observed directly in the research study. Rather, it is inferred from the relationship between the independent and dependent variable. This is an important concept to understand because most theory is based on notions of intervening variables and understanding how or why such effects occur. These variables might be clearly identified before doing a study, i.e. measured and analyzed within a study. Often, mediating variables surface as researchers interpret findings and emerge as suggestions for future research.
• Moderator Variable: a variable/characteristic that moderates or changes the direction and/or strength of the relationship between two other variables. When, under what conditions, a relationship holds; influences on the strength of the relationship. For example, if a researcher were looking at the relationship between Socio economic status and AIDs prevention, age might be a moderator variable such that the relationship is stronger for older kids than younger kids.
Understanding the distinction between mediators and moderators is not always easy. Basically, in a mediation model the independent variable cannot influence the dependent variable directly and does so by means of another variable – the mediator. As a simple example, older people tend to be better drivers than young people. So, age is a predictor of good driving. However, when we think about why this is the case, we see that older people typically make wiser decisions and so wisdom could be seen as the mediating variable.
There are a number of tests that can be used within your statistical software program to test for mediating and moderating effects. Moderated regression is an example. A moderator analysis is used to determine whether the relationship between two variables depends on (is moderated by) the value of a third variable. You can find to explore how this is conducted for the statistical package you are using. effect.