How to make measurements more accurate: (need help) (in Off-topic)


AdminShade March 5 2006 3:09 PM EST

Ok I have been doing some research on some spells but I don't know if I can call certain measurements accurate.

for example Here some measurements which I will not clarify more than needed:

Amount of measurements: 25 data points

Lowest: 108
Highest: 203
Average: 159,4
Standard Deviation: 32,79

I know the Standard Deviation depends on the measurements but what would be a way to make it smaller?

Would I make it more accurate when making 50 data points?

Is that standard deviation acceptable?

If anyone could shed some light on this 'Shadey' figures then I'd be happy :)

QBOddBird March 5 2006 3:11 PM EST

Correct, more data points = greater accuracy.

AdminShade March 5 2006 3:13 PM EST

Some more data for you people to work with / look at:


Lowest: 108 119 133 155 156 153 271
Highest: 203 234 261 292 290 244 288
Average: 159,4 175,4 188,7 218,2 224,2 199,3 279,5
ST Dev: 32,79 35,56 37,14 45,21 42,04 31,32 12,02

stabilo March 5 2006 3:16 PM EST

I think that the Standard Deviation will stabilize at some point (maybe 32,79) no matter on how many data-points you'll add. But I don't know if this is after 25 or 250 data points, you should just try this out for once, and then stick to this in all your tests you are going to do.

This will be the random-factor jon put in place, and this would probably be a very accurate calculation of this factor ;-)

regards, and I'm curious about your results Shade!

AdminShade March 5 2006 4:33 PM EST

stabilo: why do you think this has got anything to do with CB? It could be from something irl...

but yes it has to do with CB testings, and the "Jon factor" is something which isn't discussed here.

^_^

Sacredpeanut March 5 2006 5:51 PM EST

There is no way to reduce the standard deviation of the points in the sample. However you can reduce the standard error of the mean.

What is meant by this? If you were to take another sample of 25 data points you would probably not get a mean of 159.4, the standard error of the mean is basically the standard deviation of the means of your sample, and is calculated by Standard Deviation/square root of the number of data points.

This means your original sample mean has a standard error of around32.79/sqr(25) = 32.79/5 = ~6.5. In other words if you to take heaps and heaps of samples of 25 points the means of each of the samples would have a standard deviation of 6.5.

If you multiply this number by two you can get the margin of error of your sample mean - i.e the margin of error of your 159.4 value is 159.4 +/- 13. This means that the actual mean of whatever you are trying to find out lies between 146.4 and 172.4 (at 95% confidence).

Is this "accurate" enough? It depends on what you're trying to find out, if you were trying to find whether a base MH does more damage than a base BoTH for example and your sample gave a mean damage for the MH of 100 +/- 20 and a mean damage for the BoTH of 105 +/- 20, you couldn't say for sure that the BoTH does more damage since there is significant overlap between the two margin of errors.

To decrease the margin of error simply increase the sample size, increasing the sample size fourfold will half the size of your margin of error.

Stephen March 5 2006 7:03 PM EST

Shade, your opening sentence stated "I have been doing some research on some spells", so unless you are a practicing wiccan it's fair to assume you are referring to CB related issues!

Doesn't standard deviation assume a bell curve distribution of points? Are you making this implied assumption, or could your data points be randomly spread between your low and high as bounds?

AdminShade March 6 2006 12:56 AM EST

Well the data points are randomly divided between highest and lowest i think, and that mean error was exactly what i was looking for.

"What is meant by this? If you were to take another sample of 25 data points you would probably not get a mean of 159.4, the standard error of the mean is basically the standard deviation of the means of your sample, and is calculated by Standard Deviation/square root of the number of data points. "


should be going to work on this tonight also then:)

stabilo March 6 2006 2:51 AM EST

Shade: "why do you think this has got anything to do with CB? It could be from something irl... "

Sure it could, I assumed it because of those postings prior to this one:
Need some help with testing things:(look inside)
WTB Base ammo
WTB: Useless ~250k mpr single minion character

regards

AdminShade March 6 2006 3:41 PM EST

Ok increased the measurement points from ~25 to ~50 (some deviation due to randomnesses)

As you, and I of course, can see, the 'margin of error' has been decreased for most sets. latest set hardly counts because of only 2 measurements.

Standard Deviation: 32,79 35,56 37,14 45,21 42,04 31,32 12,02
Margin of Error: 13,12 14,23 14,85 18,08 16,81 15,19 17,00

Standard Deviation: 32,79 35,56 37,14 45,21 42,04 31,32 12,02
Margin of Error: 9,28 10,06 10,50 12,79 11,89 11,07 17,00
This thread is closed to new posts. However, you are welcome to reference it from a new thread; link this with the html <a href="/bboard/q-and-a-fetch-msg.tcl?msg_id=001jRv">How to make measurements more accurate: (need help)</a>