Replication and drift

Week 12Due 27 May 2027Worth 25%

Twelve hundred words. Take one protocol you have already run, run it again without changing anything, and explain why the two answers differ. The explaining is the assessment: the second run is half a day of work you have already done once, and the account of the gap between the two answers is the rest of it, which is why forty-five of the marks sit there and none of them sit on the two answers agreeing.

Do not improve the protocol#

The instinct will be overwhelming. You now know more than you did in March, you can see three things wrong with how you measured it, and fixing them would obviously give a better result.

Fix nothing. A modified re-run answers a different question and tells you nothing about drift, and the marks for this piece sit in the comparison, which a modified protocol destroys.

If the original protocol was genuinely unsafe or unrunnable, say so and stop. That is a legitimate finding and it is assessed as one: name the line in the protocol that fails, say what it would take to make it safe, and say plainly that the repaired version is a different protocol and cannot be set beside the first run. Every week of this course above the lowest risk published a stopping rule and something else to do instead before it asked you to do anything, and this piece keeps the same promise.

Do not average them#

Two results, side by side, both visible. The average of a March answer and a May answer is a number describing nobody, not the person you were in March, not the person you are now.

The largest criterion, again, is the account#

You need at least one explanation for the difference that is not a mistake you made. Looking for errors first is correct, and you should. But after you have looked and found nothing, the explanations that remain are the interesting ones: the season changed, your workload changed, you have spent a semester learning to watch yourself and that changed you too. Week 4 told you the observer was in the room. Week 10 told you the subject keeps moving.

A submission that explains the whole gap as sloppiness has not understood the course. A submission that explains none of it as sloppiness has not checked.

What you are left with#

The last part of this piece, and the last thing you will write for the course, is one paragraph: the expiry date on your own main finding. When would you need to run this again before you trusted it? Say a date and say why. There is no correct answer, and there is a difference between a considered one and a shrug: a date resting on something that actually moves, the end of semester, the turn of the season, the next time your weeks look the way these ones did, is a considered one, and a date chosen because it is a round number away is not.