+x+x+xHello,
In the experiment attached, there is an event (landslide) in a condition of "extremely low probability" that is probabilistically determined to occur in 1% of the cases. However, in 660 trials (testing multiple participants, it did not happen once - do you have any idea why? For the other variables, the frequency of occurrence of the landslide is close to what we had set (10%, 33% and 66%).
I attach the programme here - it shows:
<list extremelyLowProbability_evacuate>/ items=(trial.Evacuate_NoLandslide,trial.Evacuate_Landslide)/ poolsize=10/ itemprobabilities = (.99,.01)</list>
Has the poolsize anything to do with that?
In total the experiment run 12 trials per conditions, but I would expect that in 660 (running the experiment on multiple participants) I should to see at least one landslide for this condition.
Many thanks
Elena
> Has the poolsize anything to do with that?
Yes.
<list extremelyLowProbability_evacuate>
/ items=(trial.Evacuate_NoLandslide,trial.Evacuate_Landslide)
/ poolsize=10/ itemprobabilities = (.99,.01)
</list>
A poolsize of 10 means there will be ten items in the list. With the probabilities set as they are, they'll all be "no landslide". To have at least one "landslide" item in the <list>, you would need a poolsize of 100.
Hi Dave,
Thank you for your reply. The total number of trials for that condition is 12 per experiment, but the academic wants to run the experiment on 55 participant. There will be in total 660 trials for that conditions. If she wants to calculate the probability on the total amount of trials would it be more correct to consider the total amount of trials or if wouldn't make any difference (as long it will be a multiple of 100)? As a general rule would it be more correct to consider as pool size the total number that will allow to "pick up" an integer number of event? like for instance if my probability would be 0.013 would it make more sense say that the pool size will be 1000 even if the number of trials for that condition in the experiment are 12? Would it make any sense using larger numbers that reflect the total amount of trials that I wanted to run for that condition (e.g., 660)?
Thank you,
Elena
What
<list extremelyLowProbability_evacuate>
/ items=(trial.Evacuate_NoLandslide,trial.Evacuate_Landslide)
/ poolsize=100
/ itemprobabilities = (.99,.01)</list>
does is create a list with 100 items. With the probabilities set to .99 and .01, the end result is a <list> containing
- 99 x trial.Evacuate_NoLandslide, and
- 1 x trial.Evacuate_Landslide.
You then sample from that list _without replacement_, i.e. you can think of it as a deck of 100 cards (99 cards say "no landslide", 1 card says "landslide") and a participant then draws 12 cards (=each draw is a trial) from the deck without putting those cards back. I.e., the probability that participant X will encounter the "landslide" card will change with each trial. It is
1/100 for the 1st draw/trial,
1/99 for the 2nd draw/trial,
1/98 for the 3rd draw/trial,
and so forth.
If you wanted a constant 1/100 chance per draw, you would sample _with_ replacement, i.e.
<list extremelyLowProbability_evacuate>
/ items=(trial.Evacuate_NoLandslide,trial.Evacuate_Landslide)
/ poolsize=100
/ itemprobabilities = (.99,.01)
/ replace = true
</list>
Each participant receives such a deck, i.e. whatever cards participant A draws from her/his deck has no effect on participant B's deck, they are completely independent. Hope this clarifies.