To find causal relationships between two variables, we need to design an appropriate experiment that controls for all possible confounding variables. Then we change the variable of interest to see its effect. This, in practice, is very difficult. So in statistics, we have what is called Causal Inference. This is a field that seeks to examine observational data and identify causal relationships, rather than merely correlations. How do we do this? We aim to control for all variables by matching on the most similar data points and only looking at the dependent variable. Once we do that, we will have two groups, a control and a treatment group, then all we need to do is a t-Test to measure the difference between the two groups. So how do we do that?
First, we run all of this on your device. We never send data to a server. (I don’t know how to do that; I might need one-on-one time with AI to figure that out.) We review your health data and select one variable to test. We split that variable into high and low groups based on your historical median. Using that information, we use the other variables to predict whether a data point is in the treatment group. We use logistic regression to find the probability that each point belongs to the treatment group. This probability is called the Propensity Score. Then, using that score, we match an above-average day with a similar below-average day. We use nearest neighbors for this. If a data point has no match, we drop it from the dataset.
Now, we have two groups: a treatment group and a control group. We calculate the group means. The difference between the means is the average treatment effect. We perform a two-sample t-test to test for differences in means. This test gives us the effect and its size.
If it passes, this becomes a Discovery in your app.
For the Connected States, we try to answer your body’s feedback loops. For example, good sleep gives you energy to work out. When you work out, it helps you sleep better. The cycle continues. We answer this by checking if the relationships go both ways in our previous results. If they do, it appears in your Connected States Page.
Finally, if you ever see any issues or have questions, feel free to reach out to me.
As always, I am not a doctor, and these are not 100% accurate. We calculate statistical information based on historical data. This app is not a medical device. These insights are not intended to diagnose, treat, cure, or prevent any disease. Always consult with a qualified healthcare provider before changing your health, diet, or exercise routines.