What are correlations?

Correlations describe a relationship between two attributes, if there is one. Correlations can't determine the cause of the relationship—this is up to you to have a think about! One attribute changing might cause the other to change or vice versa, or a third variable might affect both attributes, or the relationship might be a coincidence. All a correlation can tell us is that when the values of one attribute increase or decrease, the values of another attribute usually increase or decrease as well.

If both attributes increase, that's a positive correlation. For example, if you tend to rate your mood more highly on days when you also walk more, then there's a positive correlation between your mood and steps attributes. If your mood goes up when you walk less, that's a negative correlation, because the values of one attribute go down when the values of the other attribute go up.

Screenshot of a correlation in Exist that says

In Exist, each correlation includes indications of strength and confidence. The strength indicator describes how strong the relationship between the two attributes is. If your steps value is sometimes higher when your mood value is higher, the relationship is not very strong. If both values are almost always higher at the same time, then the relationship is very strong.

The confidence indicator shows a number of stars out of 5 to describe how sure we are about the relationship between these two attributes. You might have a very strong relationship between two attributes, for example, with a very low confidence if we don't have much data to use in our calculations. As you track more data in Exist over time, that correlation might become a weaker relationship, or it might stay the same but with a higher confidence, because more data points back it up over a longer period.

All correlations can be read with either attribute first, because they don't imply any causation, only that both attributes happen to go up or down on the same days. It might sometimes make more sense to you to imagine the order of attributes in a correlation swapped around, but the correlation and its strength are valid either way around, because the relationship is the same.