What statistical tests for headache journal? I track my pain levels in an online spreadsheet along with daily habits and trigger events. I want to test whether changes in my pain over time follow a trend (not concerned whether it is linear or not, just whether it is going up or down). I also want to test which daily events and triggers can be significantly associated with a change in pain level. It's been a while since I took statistics class and I'm confused by online flowcharts. Which statistical tests should I use and should my data be considered parametric or non-parametric?
Thanks.
 A: The simplest thing to do is use OLS (parametric), which estimates $\beta_i$ in the equation 
$$Pain_t=\beta_0+\beta_1 DailyEventA_t+\beta_2DailyEventB_t+\beta_3 Pain_{t-1}+\epsilon.$$
So $Pain_t$ is you pain level on day $t$, on some fixed scale (10 is too much, 5 might be reasonable), $DailyEventA$ is whatever measure you take to alleviate the pain. If it is binary (do it/don't do it), then $DailyEventA=1$ or $DailyEventA=0$. $Pain_{t-1}$ is the pain level the day before; if you assume pain is independent from day to day, you can drop this part from the equation. If you assume even pain from two days ago has an impact, you might even want to include $Pain_{t-2}$.
Why do this? It helps you to see whether $DailyEventA$ or $DailyEventB$ tends to reduce the pain level. Suppose you estimate $\beta_1=-1$ in the above equation and $DailyEventA$ is binary. That means doing $DailyEventA$ on day $t$ reduces the pain level by 1 on your scale on average that day. Run a t-test on these coefficients to determine whether the effects are statistically different from zero.
You can, however, only determine correlations with this method. If you would randomize whether you take $DailyEventA$, say throw a coin each day and according to the outcome do it or not, then you can better determine the actual effect of $DailyEventA$.
If you just want to see whether you pain level goes up or down over time, just plot the time series. Or, more sophisticated, you could estimate something like
$$Pain_t=\beta_0+\beta_1 t+\epsilon$$
to get a linear time trend. If $\beta_1>0$, then pain is increasing over time. You can also estimate nonlinear time trends by including polynomials
$$Pain_t=\beta_0+\beta_1 t+\beta_1 t^2+\beta_1 t^3+\epsilon,$$
or include as above $DailyEventA$ etc to "account for" (control for) the effect of your treatments.
To test robustness of your result, you can also use Ordered logit instead of OLS, which usually works better if the dependent variable (pain level) is ordinal rather than cardinal, which should be the case here.
PS: data is not considered parametric or non-parametric; models are.
