Two days ago I explained why the test I was using to measure my mental function many times/day had room for improvement. I wanted a new test much like the old test but with which my accuracy was higher.

I was more accurate with the simple arithmetic test (e.g., 3 + 6) than with the memory test I described two days ago. The crucial difference might have been the number of possible answers. The arithmetic test had 40-odd possible answers; the memory test had 2 (yes and no). Saul Sternberg did a reaction-time experiment in which the number of possible answers was varied from 2 to 8. I don’t know what the accuracy data were but the variance of the reaction times was lower with 8 possible answers even though reaction times were longer. A plausible explanation is that there was much more anticipation with 2 possible answers than with 8. Anticipation can cause errors.

The new test I am trying consists of typing how many letters from the set {A, B, C, D} are among a set of four letters chosen from a much larger set (most of the alphabet). The possible answers are “1″, “2″, “3″, and “4,” each equally likely. For example, I might see T B X A. The correct answer is “2″. I am using R (the programming language) to run this test so I type “2″ with one hand and hit Enter with the other as fast as possible.

Here are the results so far from the new test — the training phase.

These values are taken from fit of a linear model; they are similar to means. As I gain experience with the test I am getting faster. The new test is slower than the old test (which is good — more mental processing).

Consistent with what Sternberg found, variation in reaction times is less with the new test than with the old test even though average reaction times are greater:

This graph shows the standard deviation of residuals from the fitted model. The units are reciprocal seconds (x 10) because I did a reciprocal transformation before fitting the model. The reciprocal transformation made the reaction times close to normally distributed.

Here is accuracy:

The new test feels easier than the old test, but so far there is little difference.

Overall it seems to be a step in the right direction. Reduction in variation of reaction times means more sensitive measurements.

The experiments I am planning are very simple: Test myself regularly (say, every half-hour), eat something. If the measurements are steady, it is very easy to see an effect. As far as I know, such experiments have never been done. One reason, I think, is that they require self-experimentation: It is no trouble for me to do the test (which takes 4 minutes) 100 times in a week and thereby reach a steady state. But to have someone else do the test 100 times as preparation — especially if the test were done in a lab — would be very difficult.

During a trip to Los Angeles a few weeks ago, I noticed that my scores on several mental tests were better all of a sudden. The apparent cause was that I had taken flaxseed oil at an unusual time. Normally I took it about 10 hours before the tests; in this case I had taken it about 4 hours before.

Does flaxseed oil have a short-lived effect on brain function? When I got home I tried to find out. Rather than doing a set of four tests once per day I switched to one test many times per day (e.g., 10 times). This would allow detection of short-lived ups and downs in my mental function.

The test I used required nothing but my laptop. I usually have my laptop with me so such a test is much easier to do throughout the day than a task that requires other equipment. The test consisted of four blocks of 50 trials each. For each block I memorized a new set of three digits (e.g., 0 1 7). On each trial I saw 1, 3, or 5 digits and pressed a keyboard key as quickly as possible to indicate if any of the memorized digits is in the displayed set. For example, if the memory set was 0 1 7 and the display set was 3 2 8 the correct answer was “no” (which I indicated by pressing “4″).

The trials were packed together as closely as possible: As soon as I answered, the next set appeared. It took about 3 minutes to do 200 trials.

I did frequent measurements for four or five days. They appeared to confirm what the Los Angeles measurements suggested: Flaxseed oil did have a short-term effect. But two things muddied the water:

1. Baseline measurements were not always as steady as I would like. There were ups and downs that seemed too large to be random variation. The curious and exciting thing was that these ups and downs usually had a possible explanation — something had changed. For example, the measurements would be X1, X2, X3, Y. X1, X2, and X3 are close; Y is quite different. Between X3 and Y I had eaten a meal.

2. The task was difficult. I was about 88% correct and it was hard to do better. With any reaction-time task there is a speed-accuracy tradeoff: If you are slower, you can be more accurate. In this particular case this is a problem because it is an added source of variation and may reduce reaction-time differences: Rather than becoming slower, I become less accurate (or rather than becoming faster I become more accurate).

Problem #1 is easy (if slightly unpleasant) to solve: Keep the situation more constant. Eat less during the measurement period, etc.

To reduce Problem #2 I am learning a new task. I will go into detail tomorrow.