I posted a few days ago about the different effects of flaxseed oil (high in omega-3) and olive oil (low in omega-3) on my balance. There was a big difference. If omega-3 affects one measure of brain function (balance), it should affect many other measures of brain function. The whole brain is made of the same stuff (neurons, etc.).
Which brain measures are most sensitive to omega-3? The more processing/time the better, I assumed; so I looked for tasks that, like balance, involve continuous processing for most of the test period. This led me to try a paper-and-pencil version of Saul Sternberg’s memory-scanning task. (Sternberg’s use of this procedure is described here.) On each trial I memorized a list of three digits (e.g., 2, 3, 7); then as fast as possible marked each of 100 digits (20 digits/row in 5 rows) according to whether they were in the list or not. I made a line under the digit if it was in the list, through the digit if it was not. I did five trials per day.
Here is an example of the test materials and my marks:
The other side of the page had two more sets of digits.
Here are the results from the same flaxseed/olive oil experiment I discussed a few days ago:
There was a huge difference between the flaxseed oil and olive oil condition: t > 7.
Curiously the time course is different from the balance results. In the case of balance, when I switched from flaxseed to olive oil my balance slowly got worse. Nothing like that is apparent here. This might reflect a different mechanism or it might be due to the vast difference in how much practice I had had with each task. When this experiment began, I had had far more practice with the balance task than with the memory-scanning task.
Hi Seth,
Very intriguing results! I’ve been looking at these and other recent plots on your blog, and was especially interested in the discussion about whether a trend (or change in trend) is “obvious” as opposed to requring a more formal statistical test. One thing that should be mentioned is the effect of serial correlation on both observed visual patterns and statistical tests. A hallmark of time series data is that observations tend to be correlated over time: If I know your score today I can predict with better than chance accuracy your score tomorrow. The result is that observations are not independent, yet standard versions of statistical tests such as t-tests from linear models require this. The model parameter estimates are fine, but their standard errors are underestimated in the presence of autocorrleation, and thus statistical “significance” is overestimated.
I digitized this graph and did a quick analysis modeling the autocorellation along with the experimental condition, and I get a t-value about half of your reported t=7, depending on the exact model. Still big, so it may seem like I’m being pedantic here, but in other cases autocorrelation effects can be really important. Here are a couple of interesting links, the first to a general discussion of autocorrelation with some examples of visually striking patterns that are nonetheless random, and caused by high autocorrelations:
https://socrates.berkeley.edu/~epsc120/Toolkits/Toolkit_11.pdf
and this, an empirical study of people’s judgements about intervention effects in single-case AB designs: https://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1286245
In any case, your blog is really interesting and I’m looking forward to more data!
That’s a good point. These data do not look like what I expected nor like other self-experimental data I have collected and now that you mention it I realize that auto-correlation would explain it. I imagine it will go away with more practice.
I just had some heart news and need to go an a strict diet when I recover from Pnemonia which I have had for two or more weeks. Any thing you can tell me about which oil and which foods would help me with heart issues, high choles.. etc. is helpful. I have never been ill or had any problems and want to be very pro active with diet and exercise. Thanks, Elaine