Column Graph

Developing Column Graphs

A column graph can and should provide an easy-to-interpret visual representation of a frequency or percentage distribution of a single (or multiple) variable(s). Column graphs present a series of vertical equal-width rectangles, each with a height proportional to the frequency (or percentage) of a specific category of observations. Categories are labeled on the x-axis (the horizontal axis), and frequencies are labeled on the y-axis (the vertical axis). For example, a column graph that displays the partisan distribution of a single session of a state legislature would consist of at least two rectangles, one representing the number of Democratic seats, one representing the number of Republican seats, and perhaps one representing the number of independent seats. The x-axis would consist of the “Democratic” and “Republican” labels while the y-axis would consist of labels representing intervals for the number of seats in the state legislature.

When developing a column graph, it is vital that the researcher present a set of categories that is both exhaustive and mutually exclusive. In other words, each potential value must belong to one and only one category. Developing such a category schema is relatively easy when the researcher is faced with discrete data, in which the number observations for a particular value can be counted (i.e. nominal and ordinal level data). For nominal data, the categories are unordered, with interchangeable values (e.g., gender, ethnicity, religion). For ordinal data, the categories exhibit some type of relation to each other, although the relation does not exhibit specificity beyond ranking the values (e.g., greater than vs. less than, agree strongly vs. agree vs. disagree vs. disagree strongly). Because nominal and ordinal data are readily countable, once the data are obtained, the counts can then be readily transformed into a column graph.

With continuous data (i.e., interval- and ratio-level data), an additional step is required. For interval data, the distance between observations is fixed (e.g., Carolyn makes $5,000 more per year than John does). For ratio-level data, the data can be measured as in relation to a fixed point (e.g., a family's income falls a certain percentage above or below the poverty line). For continuous data, the number of potential values can be infinite or, at the very least, extraordinarily high, thus producing a cluttered chart. Therefore, it is best to reduce interval- or ratio-level data to ordinal-level data by collapsing the range of potential values into a few select categories. For example, a survey that asks [Page 196]respondents for their age could produce dozens of potential answers, and therefore it is best to condense the variable into a select few categories (e.g., 18–25, 26–35, 36–45, 46–64, 65 and older) before making a column graph that summarizes the distribution.

Figure 1 1990–2009 Illinois House Partisan Distribution (Column Graph)

Source: Almanac of Illinois Politics. (2009). Springfield: Illinois Issues.

Multiple Distribution Column Graphs

Column graphs can also be used to compare multiple distributions of data. Rather than presenting a single set of vertical rectangles that represents a single distribution of data, column graphs present multiple sets of rectangles, one for each distribution. For ease of interpretation, each set of rectangles should be grouped together and separate from the other distributions. This type of graph can be particularly useful in comparing counts of observations across different categories of interest. For example, a researcher conducting a survey might present the aggregate distribution of self-reported partisanship, but the researcher can also demonstrate the gender gap by displaying separate partisan distributions for male and female respondents. By putting each distribution into a single graph, the researcher can visually present the gender gap in a readily understandable format.

...