This past week (I am not certain how to define week anymore) the course focused on bivariate statistics. That is, the nature of the relationship of two variables. Or, a more precise "that is"- does a change in one variable lead to a change in the other? Does smoking more cigarettes lead to a higher chance of causing something (quantitative) and does being female result in being smarter (qualitative)?
This ties in nicely with statistical significance. Haaf et al, Retire significance, but still test hypotheses really want to be done with significance. So much so, that they got 800 scientists to sign that they also want to be done with significance. Significance means that the results are more than just chance. Or so I thought.
In the article, they blame scientists for using significance as a kind of pass/fail criteria. As a result of significance being the end-goal, they claim scientists often massage data and cherry pick methods so as to prove significance. Finally, the arbitrary level of 0.05 shouldn't be used blindly.
The solutions they propose are: pre-register studies and print all results, don't be so confident and categorical but rather try to think through the data and realize uncertainty is built in (they reccomend "compatible intervals" rather than confidence intervals), calculate a p value accurately rather than blindly using 0.05 and be humble.
As scientists are known for their ability to be humble, I doubt the last recommendation should be difficult. And I really do agree that studies should be pre-registered. But the big problem is as follows: significance serves to add objectivity and steer clear from intuition. Of course it would be wonderful if people were thoughtful and clever and objective. And while many people can be thoughtful, really are clever, and try to be objective, I doubt significance is the problem. After all, if those who used significance incorrectly weren't able to learn what significance (a relatively not challenging term) means, why would they take the time to do more complicated statistics? Intuition is a big problem (as the reproducibility problem proves), and weeding out significance is just a cosmetic solution.
I realize non of this is physical and all of this is data... Oh well.
Saturday, 13 April 2019
Sunday, 7 April 2019
Statistics and Compliance, II
This week (if a week is defined as "since last time I posted") I learned that in case of a normal distribution, the mean/standard deviation is to be used. However, if the data is skewed, median/five number summary is to be used. I expect the latter (where median is the middle data) is better in skewed populations since it will be less affected by extreme cases. However, if you have two studies, and one gets results that are skewed while the other has a nice, normal distribution- how can you compare?
I realize this: I have never, ever ever ever seen the median being used to describe a population. I would say "and IQR even less so" but you can't have less than zero. Weirdly, I HAVE seen articles where the mean-standard deviation gives a value less than zero for something that can't be less than zero (pain scale, for example).
Fortunately, the article I read was great for continuing to wonder how to compare articles, and also continues the article from last week (again, week meaning last blog post). The article is Barriers to Treatment Adherence in Physiotherapy Outpatient Clinics: A Systematic Review by Jack et al.
The article narrows hundreds of potential articles to 20 cohorts and aims to identify barriers to adherence.
The punchline is that most studies found a low level of physical activity predicted low adherence to physical activity. Then a whole bunch of psychological reasons (depression, anxiety, etc), low social support and finally pain. The place of psychology in physical therapy/exercise is fascinating, as CBT and mindfulness courses seem to be creeping. But back to the main point. If you don't exercise, you probably won't exercise. I can't decide if that's wholly obvious to the point of silly, or really depressing to the point of depressing. We should start paying people to exercise.
And now for the technical aspect of comparing wildly different studies. What a fun mess! How do you define adherence? What is the lack of adherence percentage? Not known. Though the latter is anywhere between 14$-70% or above or below. And this was just in the introduction! The studies are so completely different in population type, size, and methods. Population type- they examined very different types of physical therapy such as pelvic floor, sports injury and osteoarthritis. The populations size in the studies varied extremely by size; one had 34 participants. Most had more.The methods were mostly questionnaire, but obviously different questionnaires, and thus comparing statistics really isn't quite right. Any guesses on how many studies used the median?
I realize this: I have never, ever ever ever seen the median being used to describe a population. I would say "and IQR even less so" but you can't have less than zero. Weirdly, I HAVE seen articles where the mean-standard deviation gives a value less than zero for something that can't be less than zero (pain scale, for example).
Fortunately, the article I read was great for continuing to wonder how to compare articles, and also continues the article from last week (again, week meaning last blog post). The article is Barriers to Treatment Adherence in Physiotherapy Outpatient Clinics: A Systematic Review by Jack et al.
The article narrows hundreds of potential articles to 20 cohorts and aims to identify barriers to adherence.
The punchline is that most studies found a low level of physical activity predicted low adherence to physical activity. Then a whole bunch of psychological reasons (depression, anxiety, etc), low social support and finally pain. The place of psychology in physical therapy/exercise is fascinating, as CBT and mindfulness courses seem to be creeping. But back to the main point. If you don't exercise, you probably won't exercise. I can't decide if that's wholly obvious to the point of silly, or really depressing to the point of depressing. We should start paying people to exercise.
And now for the technical aspect of comparing wildly different studies. What a fun mess! How do you define adherence? What is the lack of adherence percentage? Not known. Though the latter is anywhere between 14$-70% or above or below. And this was just in the introduction! The studies are so completely different in population type, size, and methods. Population type- they examined very different types of physical therapy such as pelvic floor, sports injury and osteoarthritis. The populations size in the studies varied extremely by size; one had 34 participants. Most had more.The methods were mostly questionnaire, but obviously different questionnaires, and thus comparing statistics really isn't quite right. Any guesses on how many studies used the median?
Subscribe to:
Posts (Atom)