Row Conditional Distributions

From Counts to the Composition of a Row

A two-way table organizes counts according to two categorical variables. Its cell counts tell us how many observations fall into each combination of categories. But a count by itself can be hard to interpret when the row categories contain different numbers of observations. A row conditional distribution changes the scale: within each row, it reports the proportion of observations in each column category.

For example, suppose a hypothetical table records whether appointments at several clinics were completed or missed. If one clinic's row contains 30 appointments and another's contains 40, comparing their raw counts of completed appointments does not directly describe the composition of either clinic's appointments. Dividing each cell count by its own row total answers a focused question: among the observations in this row category, what fraction belongs to each column category?

How to Calculate a Row Conditional Distribution

Suppose the rows represent categories of one variable and the columns represent categories of another. Let \(n_{ij}\) be the count in row \(i\), column \(j\), and let \(n_{i+}\) be the total count in row \(i\). The plus sign indicates that the counts across the columns have been added for that row. The row conditional proportion for cell \((i,j)\) is:

To express that proportion as a percentage, multiply it by 100. The numerator is one cell count; the denominator is the total for the same row. Calculate each row separately. A row conditional distribution is not the same as a distribution of proportions out of the entire table: its denominator is the row total, not the grand total.

The table must have meaningful counts for each row. In particular, every observation included in a row total must be classified into exactly one of the column categories. If there are unclassified or missing observations, the analyst must decide how they are handled and make clear what population the row total represents. If a row total is zero, no row conditional proportions can be calculated for that row, because division by zero is undefined.

Why Each Row Adds to One

A row conditional distribution should account for the entire row. If the column categories are exhaustive and mutually exclusive for the observations in that row, their cell counts add up to the row total. Dividing each of those counts by that same total makes the row proportions add to one.

This is a short algebraic justification for the row-sum check. The numerator \(\sum_j n_{ij}\) is the sum of all cell counts in row \(i\), which equals \(n_{i+}\). Since a positive row total divided by itself is 1, the row proportions sum to 1. If reported as percentages, they sum to 100%, apart from small discrepancies caused by rounding.

The check is useful, but it does not by itself prove that every calculation is correct: an incorrect set of values can sometimes still add to one. Check both that each cell was divided by its own row total and that the resulting row sum is one.

Presenting and Interpreting the Results

A clear presentation keeps the row categories and column categories visible, labels the entries as proportions or percentages, and states what the percentages are percentages of. One convenient format is a table with each row converted to percentages. For the travel-mode example, that format is:

A useful interpretation names the row category and the column category. For example: “Among the surveyed students living on campus, 54.2% reported that they usually take the bus.” This states the relevant group, the measured category, and the percentage's denominator in words. Avoid writing only “54.2% take the bus,” which leaves unclear which group the percentage describes.

It is often helpful to show one decimal place for percentages, but the level of rounding should suit the size and precision of the data. Keep enough digits during calculation, then round the displayed result. Because individually rounded values may add to 99.9% or 100.1%, a small discrepancy from 100% is not necessarily an error. The unrounded proportions should sum to 1.

Common Mistake: Using the Wrong Denominator

A frequent error is to divide a cell by the grand total of the table instead of by its row total. That produces a joint relative frequency: the fraction of all observations in the particular cell. It does not give the composition of the row. For instance, in the appointment table, \(18/100=18\%\) is the share of all 100 appointments that were both from the North clinic and completed. The North row conditional proportion is \(18/30=60\%\), the share of North clinic appointments that were completed. The two values answer different questions.

Another error is to divide by a column total. A cell divided by its column total describes a proportion with a different reference group, so it is not the row conditional proportion. For this topic, keep the question in view: within this row category, how are the observations distributed across the columns? The answer uses the row total as the denominator.