Some of us here are having problems understanding
correlation.  What is the best way to explain this to people
who might be an expert in healthcare, but not a statistician?

Response:

Correlation is a statistical technique which can show whether and how strongly pairs of variables are related.  For example, look at the amount of rainfall on a given day and the number of people with umbrellas.  You’ll certainly find that rainfall increases the number of people with umbrellas.  However, if you looked at the amount of water on the ground (say, after a major storm just came through), you’ll find something different.  While rain can be shown to cause an increase in umbrellas, just measuring how much water is on the ground doesn’t tell you much.  Wet ground does not lead to umbrellas unless rain is falling.  So while rain and umbrellas are correlated, water on the ground and umbrellas do not have a direct correlation.  Water on the ground and umbrellas are correlated to rainfall—but not to each other.

For a healthcare example, let’s consider survey questions that focus on the cleanliness of the room, the kindness of the nurse staff, and the quality of the cafeteria food.  While there is a substantial body of research to support that the first two are highly correlated to Overall Patient Satisfaction, the quality of the cafeteria food is rarely a driver of Overall Patient Satisfaction.  It doesn’t take much imagination to see that if the nurses were kind and the room was clean, the cafeteria’s importance pales by comparison.

An important aspect of correlation is that a high correlation score does not always prove a causal relationship.  For example, height and weight are correlated—tall people are, on average, heavier than short people.  However, being heavy does not cause tallness (wouldn’t that be nice!), and being tall doesn’t necessarily cause heaviness.  Height and weight are simply two variables that tend to rise and fall with each other.

Correlation is the key driver of HealthStream Research’s Action Plans and recommendations to clients.  Focusing on the attributes that are highly correlated to Overall Satisfaction gives you the greatest return on investment. In many ways, our recommendations are like data “triage.”  When a patient arrives in critical condition, you know to focus on the problems that have the highest correlation to life.  If the patient isn’t breathing and has a broken toe, it’s pretty easy to see where to start first!  The same goes for survey results.  If your hospital scores high on every attribute except nurse kindness, you’ll probably find that your Overall Satisfaction decreases.  This is because nurse kindness is one of the strongest drivers (i.e. has a strong correlation) of Overall Satisfaction.

The main result of a correlation is called the correlation coefficient (or "r") and ranges from -1.0 to +1.0. The closer “r” is to +1 or -1, the stronger the association of the two variables.  If “r” is close to 0, it means there is no relationship between the variables. If “r” is positive, it means that as one variable gets larger the other gets larger. If “r” is negative it means that as one gets larger, the other gets smaller (often called an "inverse" correlation).