Ken Blake, Ph.D.


How to know what you're doing

Like hand tools, each statistic has a particular job that it's designed to do. Using a statistic for a job it's not designed to do is like driving a nail with a pipe wrench. It might work, but probably not all that well.

Here's a flow chart that can help you pick the right statistic for what you need to accomplish. It covers only the statistical tools described in this minicourse. There are other statistics for other situations not covered by the chart. This chart appears at the end of each video along with an explanation of how the statistical procedure the video demonstrates fits into the chart.

 

Here are some things you'll need to understand in order to navigate the chart correctly:

How to count variables

Right away, you have to figure out whether the analysis you want to do involves one variable or two. So let's look at that first.

Starting with how to recognize two-variable analyses might be easiest. In two-variable analyses, one of the variables is, or at least could be, thought of as causing the other. For example, think about the analysis looking at whether the type of campaign ad (issue ad or image ad) is related to whether the ad does or does not use a fear appeal. In the video describing how to do that analysis, I suggested that the type of ad might be a cause of whether the people who produced the ad decided to include a fear appeal. Similarly, in the two analyses of the effects of viewing a political debate, it seems reasonable that viewing the debate might cause a change it attitudes toward the candidate. These are all two-variable analyses.

By contrast, the analysis comparing the percentages of victims by race in both homicide news coverage and actual homicide statistics involves only one variable: the race of the victims in the sampled TV news stories about homicide. Yes, the actual crime statistics provide an idea of what the percentages would be if the coverage were unbiased. But there's no basis for suspecting a causal relationship between the actual crime data's proportions and the TV news coverage's proportions. The actual crime data simply provides a known standard against which to compare the TV news coverage's percentages.

Similarly, there's no good reason to think that the amount of TV that Americans watch would cause or influence the amount of TV that Canadians watch. In that example, the average amount of television that Americans watch simply provides a standard against which to compare the amount that Canadians watch. So, that analysis also involves only one variable: The amount of TV that Canadians watch.

Remember, too, that social scientists hedge a lot when calling something a "cause" and something else an "effect," because it's actually pretty easy to be wrong when you say that one thing causes another. For purposes of figuring out how many variables are involved in an analysis, you can think informally about causes and effects.

Determining measurement levels

Broadly, there are two kinds of variables: categorical ones and continuous ones.

A categorical variable specifies which one of two or more categories a thing fits into. For example, the condition of a light bulb can be classified as either "on" or "off."

Meanwhile, continuous variables describe the amount of some property that the thing has. For example, that same light bulb could be described in terms of how much light it is emitting. It could be described as emitting no light, a little light, some light, or a great deal of light. Or it could be described (more precisely) as emitting no lumens, 250 lumens, 800 lumens, or 1,600 lumens.

Some categorical variables from the examples on this website include:

1. Race of homicide victime (Black, Latino, other, or white)

2. Exposure to candidate debate (exposed, or not exposed)

3. Type of ad (issue ad, or image ad)

4. Type of appeal (fear appeal, or not fear appeal)

Meanwhile, some continuous variables from the examples on this website include:

1. Hours spent watching television per day (Could range anywhere from zero to 24)

2. Candidate approval (Could range anywhere from 1, meaning little approval, to 10, meaning a lot of approval)

3. Body fat percentage (Could range anywhere from near zero to ... well, let's hope somewhere lower than 100 percent).