Y buy points in basketball
- Thread starter verbalkint
- Start date
Just curious why you think it would take a million trials to be statistacally meaningful...I am assuming you just threw that number out there. Honestly would like an answer so I can get a read on you, no disrespect intended...
The million trials thing is wrong. If we're talking in general terms, the lowest number of trials that is "statistically significant" is 30. There is no number that can be assigned to that (in terms of confidence level, precision, or variability) but it is considered "significant". There are very complicated proofs and mathematical equations that I don't pretend to understand which prove that. The only reason I know that is that I am an auditor and we do testing on a sample basis all the time, so I've had to attend a couple of seminars on the statistics of sampling.
However, outside the general term of "statistically significant" used with 30 observations, that phrase has very little meaning. You need to assign other variables, the most common of which are confidence, precision, and variability. Without getting too technical, they are loosely defined as:
p -- Precision--how close the sample value should be to the true value (i.e. + or - 3%)
c -- Confidence--how certain that the observation is within the
given precision. (i.e. 95, 96% etc.)
v -- Variability--what percent of the sample will be in the most
important category? (i.e. 98%, 99%, etc.)
I got that from a program that we use at work. The very interesting thing is that when you run the program it turns out that the size of the population has very little to do with how many you need to use as a sample to get a very "statistically significant" result. This actually is in line with what we see all the time in the real world. Voter results are projected with sometimes only 1% of precincts reporting, and all Nielsen ratings come from less than 35 homes.
However, outside the general term of "statistically significant" used with 30 observations, that phrase has very little meaning. You need to assign other variables, the most common of which are confidence, precision, and variability. Without getting too technical, they are loosely defined as:
p -- Precision--how close the sample value should be to the true value (i.e. + or - 3%)
c -- Confidence--how certain that the observation is within the
given precision. (i.e. 95, 96% etc.)
v -- Variability--what percent of the sample will be in the most
important category? (i.e. 98%, 99%, etc.)
I got that from a program that we use at work. The very interesting thing is that when you run the program it turns out that the size of the population has very little to do with how many you need to use as a sample to get a very "statistically significant" result. This actually is in line with what we see all the time in the real world. Voter results are projected with sometimes only 1% of precincts reporting, and all Nielsen ratings come from less than 35 homes.