In the fast-moving world of computer-assisted reporting, journalists each day defy the widely held notion that reporters and numbers don't mix. They break down databases and computer record layouts with virtual ease, using programs such as FoxPro, Paradox, Access and Excel. Just when it seems like they've done it all, something new comes along something like statistics. The use of statistics for research and analysis in the social sciences and in most medical fields has been going on for years.
But it is a tool that is relatively unfamiliar to journalists, even those who are savvy in the general applications of CAR. That does not have to be the case, says Philip Meyer, author of "The New Precision Journalism" and William Rand Kenan, Jr., professor of journalism at the University of North Carolina at Chapel Hill. The power of statistical analysis can be tapped by any journalist who is willing to learn, Meyer says.
Both Meyer and John Bare, a media consultant who works closely with Meyer and specializes in the use of statistics, agree that the first step in interpreting statistical analyses is to understand the basic model on which it based - the linear model. "Understanding the linear model," Bare says, "is the key to understanding other tools like linear regression or logistic regression. You should always know it first." The linear model, Meyer says, "is based on the idea that many relationships in life fit a straight line, even when they don't seem to at first." One way to conceptualize the linear model is to think back to the old days of high school or college algebra.
On a basic graph with a y axis and an X axis, the linear model posits that as the X, or independent variable, changes, and so changes the y ,or dependent variable. In other words, the amount that y increases or decreases is directly dependent on changes in the X variable. Consider the real life application offered by Meyer, which deals with state SAT score averages. "The scores from each state would be directly dependent upon how many students took the test.
The fewer students who take it, generally that means a higher score, and vice versa," Meyer says. The degree to which a comparison fits a straight line is referred to as a linear correlation. So, for example, there is a strong linear correlation between the temperature outside and how much you sweat. Linear correlation leads to the discussion of the first and simplest type of statistical analysis journalists use.
It is called linear regression and follows the exact concept explained by the linear model. Regression refers to the relationship between two or more correlated variables. A more complicated application of statistics is called logistic regression. The main difference between linear and logistic regression is in the nature of the X and y (independent and dependent) variables. Meyer says that a good use of linear regression is for dealing with economics.
It allows you to make a prediction about something based on information you are sure about. For example, it is known that when the interest increases, the demand for housing decreases. Based on this linear correlation, we should be able to predict a rise or fall in the demand for housing by simply looking for a trend in interest rates. In linear regression the two variables are continuous, meaning that there is a great range in their potential value.
In the example of the SAT scores, the dependent variable can range from 400 to 1600, and the independent variable can be any number of students. In logistic regression, however, at least one of the variables involved is referred to as a categorical or binary variable. In stark contrast to continuous variables, categorical variables have dichotomous values such as male/female, yes/no or dead/alive. Logistic regression still involves finding a relationship between two or more factors, but it allows you to do it in a far more powerful, meaningful way.
"The beauty of logistic regression," Meyer says, "is that you are able to take several independent variables and use them to estimate the odds that your dependent variable will be one or zero." Using the different independent variables, Meyer says, enables you to have control over certain variables that you ordinarily may not. A case where this is most evident is in medical reporting. Very often reporters are interested in investigating mysterious death rates (death is a categorical variable) at a particular medical institution.
Using logistic regression allows them to control for certain independent factors such as the patient's age, level of income, health history or insurance coverage. Controlling for these factors in a comparison of death rates has the effect of leveling the playing field for all the institutions being examined. This is precisely what New York Times reporter Josh Barbanel attempted to do in his investigation of the unusually high rate of newborn deaths at certain New York City hospitals.
Multiple regression is very similar to logistic regression with the exception that you are able to include more than one dependent variable. Many education beat reporters have found this technique particularly valuable, Meyer says, when attempting to measure the efficiency of particular schools in a district by using graduate rates and test scores of students. In these studies, variables such as the amount of money per pupil and the level of income per family can be controlled for.
Dave Davis of The Plain Dealer+ found this out when he embarked upon a study of the test scores for 600 Ohio public schools. Davis and three other education reporters controlled for variables such as poverty, parents' education level, and the percentage of renters in a district as a measure of how often families move. Their study found that these factors these account for nearly 60 percent of the difference in test scores between districts.
But more importantly, they were able to target schools that were not performing up to the standards which Davis' demographic analysis predicted they should. Some schools, on the other hand were performing well above predicted levels, despite difficult social circumstances. A final tool, which Meyer says he finds very useful in making his stories relevant to their readers, is reliability analysis. It is based on the principle of intercorrelation, which measures the degree to which several different items measure the same thing.
This is particularly helpful in doing surveys. "Reliability analysis allows you to take several different survey questions and find out how related they are to each other," he says. "Once you do that, you can compute a single number, which expresses your outcome." By expressing complete survey results with only one figure, reporters can easily compare results from different segments of the population without drowning the reader in numbers.
Regardless of which method reporters decide is best for their story, the use of statistics is not as scary as it may sound. But Davis is quick to note that it is only part of the process. "It's great for getting numbers crunched," he says, "but you still have to put everything in a relevant context for your readers."