Tilted Forum Project Discussion Community  

Go Back   Tilted Forum Project Discussion Community > The Academy > Tilted Knowledge and How-To


 
 
LinkBack Thread Tools
Old 10-16-2005, 07:31 AM   #1 (permalink)
gal
Insane
 
gal's Avatar
 
Test of mean for small samples

I've tried to read up on this in my old statistics book, but most of my skills in this field are long gone..

I have 4 samples of batch Y and 2 samples of batch X. How can I test whether E(X) > E(Y)? I suspect I don't have enough observations to test this. I looked up some tables on signed-rank tests and the smallest observations they had were 4 to 5. The outcome should follow the normal distribution though so maybe I can use some other test..

Y: 10269, 16806, 14023, 8307
X: 15332, 20149
gal is offline  
Old 10-17-2005, 07:29 PM   #2 (permalink)
Crazy
 
I never took a statistical class, but I had one in probabilities. Calculate the average and the standard deviation for both X and Y. With average and the stdev you can plot a gaussian random variable for, lets say gX (for X) and gY (for Y).

The next step is something to do with ln, and then division and subtraction, and than an integration. But it is late and I cant remember. In the end you get the probability (between 0 and 1) that X is greater than Y.

I did this once in a research paper comparing the predicted resistance to the tested resistance of deep pile foundaitons. (When the predicted resistance is greater than the tested load capacity, then designed foundations could fail. The point was to evaluate the reliability of various prediction methods).
BAMF is offline  
Old 10-25-2005, 07:56 PM   #3 (permalink)
On the lam
 
rsl12's Avatar
 
Location: northern va
I'll bump this one. I'm clueless, but I'm also interested. I sometimes run into problems with small data sets myself, and would like a statistician's POV on what to do.

My uneducated guess is just to compute means and leave it at that....
__________________
oh baby oh baby, i like gravy.
rsl12 is offline  
Old 10-25-2005, 08:40 PM   #4 (permalink)
Junkie
 
sapiens's Avatar
 
Location: Some place windy
Quote:
Originally Posted by gal
I've tried to read up on this in my old statistics book, but most of my skills in this field are long gone..

I have 4 samples of batch Y and 2 samples of batch X. How can I test whether E(X) > E(Y)? I suspect I don't have enough observations to test this. I looked up some tables on signed-rank tests and the smallest observations they had were 4 to 5. The outcome should follow the normal distribution though so maybe I can use some other test..

Y: 10269, 16806, 14023, 8307
X: 15332, 20149
I seriously doubt that you would find a significant difference between the two groups regardless of what test you choose. Your sample sizes are miniscule.
sapiens is offline  
Old 10-26-2005, 06:01 PM   #5 (permalink)
Crazy
 
I have this written down on a piece of paper in my office. I will bring it home tommorow. From what I remember, the integral could not be solved analytically. So a numerical approximation is used at the end, probably gaussian quadrature. And sample size is irrelevent. All that is reqruied is the mean and standard deviation.

The biggest assumption is that the data will comply to a gaussian random variable. This is applicable for people heights, etc... However, it is not applicable for linear reandom (like rolling a dice).

And another thing, the result is not a test that X > Y. It is the probability that if you sampled X and Y one more time, the X result would be greater than the Y.
BAMF is offline  
Old 10-26-2005, 06:04 PM   #6 (permalink)
Crazy
 
Of course, the quality of the mean and standard deviation depend on the sample size. Maybe another problem would be finding the fractal dimension of the mean and stdev versus number of samples. But that was just an idea that came to me now; I do not know if there is a procedure in existance for determining how many samples need to be taken for quality data.
BAMF is offline  
Old 10-26-2005, 06:06 PM   #7 (permalink)
Crazy
 
Oh, gaussian and normal distributions are the same thing. Its the bell curve on the pdf graph.
BAMF is offline  
Old 10-26-2005, 07:51 PM   #8 (permalink)
Crazy
 
Location: U.S.A
It depends on the nature of your variables, and what assumptions you can make about the data. It looks like you have one dependant variable, and one independent variable. Your dependant variables looks like it is continuous, and your independent variable looks nominal.
It seems like an Independent Samples T-test would answer your question. This would tell you if the difference of the means of Y and of X is statistically significant. However, due to your sample size (n=2, and n=4; N=6), you will not have enough statistical power to find any differences that may exist.
lpj8 is offline  
Old 11-03-2005, 03:53 PM   #9 (permalink)
gal
Insane
 
gal's Avatar
 
Thanks all. The numbers are cycles to failure for two different batches of material, and the independend variable is the load level which is the same for all the outcomes here. The error is usually assumed to follow a normal distribution, but in this case I don't think I can say anything about the pdf. That's why I was looking into signed rank tests, but in this case I guess the sample size is too small to make any inferences.
gal is offline  
Old 11-06-2005, 09:04 AM   #10 (permalink)
Mjollnir Incarnate
 
Location: Lost in thought
You could use the t-score approximation. It's for estimating the mean of the population from small samples.

Statistically, there is a 90% chance that the population mean lies
Y: between 7879.97 and 16822.53
X: between 2533.25 and 32947.75

If you had a larger sample, it wouldn't be such a terrible range.
Slavakion is offline  
Old 11-06-2005, 12:58 PM   #11 (permalink)
Rawr!
 
skier's Avatar
 
Location: Edmontania
If you compare the average mean from X (17740.5) and the average mean for Y (12351.25) you can see that X is larger than than Y. But it doesn't really mean very much with the small sample size and huge deviation (X is 3406.13 and Y 3800.49). There is huge room for error- as you can see in slavakion's t-score.

Looking at the p-valueof a 2 sample t-test it seems to me that there's a .75 probability that x is larger than y.

Of course, i'm pretending like I know this stuff- but i'm just taking the class this semester. So take from it what you will.
__________________
"Asking a bomb squad if an old bomb is still "real" is not the best thing to do if you want to save it." - denim
skier is offline  
Old 11-06-2005, 01:13 PM   #12 (permalink)
Mjollnir Incarnate
 
Location: Lost in thought
Quote:
Originally Posted by skier
Of course, i'm pretending like I know this stuff- but i'm just taking the class this semester.
Yeah, me too. But it was good practice for my next test.
Slavakion is offline  
Old 03-03-2006, 12:23 PM   #13 (permalink)
Crazy
 
I did this recently, so im posting even though this is old.

Assuming X and Y and gaussian (normal/bell curve) Random Variables.
G = Y - X
Therefore G is a gaussian random variable.
the probability that G is less than 0 equals the probability that X > Y

Mean for G:
E[G] = E[Y] - E[X]

Var[X] = E[(X-E[X])^2] = stdev[X]^2
Var[Y] = E[(Y-E[Y])^2] = stdev[Y]^2

Var[G] = E[(G-E[G])^2]
Var[G] = E[(Y-X-(E[Y]-E[X]))^2]
Var[G] = E[((Y-E[Y])-(X-E[X]))^2]
Var[G] = E[((Y-E[Y])^2+(X-E[X])^2-2*(Y-E[Y])(X-E[X]))]
Var[G] = Var[Y] + Var[X] - 2*COV[X,Y]
If X and Y are independant then the covariance equals zero.

You now have the stdev and mean of G, and can use the characteristic function of a gaussian random variable to find the probability.

For the numbers you provided,
E[G] = mean of G = -5389.25
Var[G] = variance of G = 26045501
Stdev[G] = sqrt(var[G]) = 5103.479

Using E[G]/Stdev[G] = -1.056 and a table for characteristic function values, the probability that X > Y = 0.8531

In other words, 85%.

Correct me if you can.
BAMF is offline  
Old 03-04-2006, 09:00 AM   #14 (permalink)
Apocalypse Nerd
 
Astrocloud's Avatar
 
Quote:
Originally Posted by gal
I've tried to read up on this in my old statistics book, but most of my skills in this field are long gone..

I have 4 samples of batch Y and 2 samples of batch X. How can I test whether E(X) > E(Y)? I suspect I don't have enough observations to test this. I looked up some tables on signed-rank tests and the smallest observations they had were 4 to 5. The outcome should follow the normal distribution though so maybe I can use some other test..

Y: 10269, 16806, 14023, 8307
X: 15332, 20149

The thing that drives me nuts about these theoretical problems is that the problem is solved with more data. Right now we are talking about confidence levels of a weak correlation that Set X belonging in Set Y. How do we know that set Y doesn't represent a measure of cloud formations and Set X represent some measure of the mating patterns of sea horses? i.e. what exactly are we testing?

I have a very good friend who has a PhD in stats. I will see if he will entertain this.
Astrocloud is offline  
Old 03-04-2006, 09:35 AM   #15 (permalink)
Crazy
 
I assumed independant. Thus correlation coefficient is zero.
BAMF is offline  
Old 03-06-2006, 02:22 PM   #16 (permalink)
gal
Insane
 
gal's Avatar
 
Heh, nice to see I'm inspiring homework. I dunno about using the standard normal, BAMF. The t-test should be used for small samples, and back in the days I remember I had to use non-parametric methods for even smaller samples.. but I can't remember why.
gal is offline  
Old 03-07-2006, 12:10 PM   #17 (permalink)
Apocalypse Nerd
 
Astrocloud's Avatar
 
ANOVA test is what is typically done for normal sized samples. You MUST use the t-test but your degrees of freedom go through the roof. -from the prof.
Astrocloud is offline  
Old 03-08-2006, 01:07 PM   #18 (permalink)
Riding the Ocean Spray
 
BadNick's Avatar
 
Location: S.E. PA in U Sofa
I think statistics are against me today. I came in here to see what these tests of men with small nipples might be all about.
BadNick is offline  
 

Tags
samples, small, test

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On



All times are GMT -8. The time now is 04:32 AM.

Tilted Forum Project

Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2024, vBulletin Solutions, Inc.
Search Engine Optimization by vBSEO 3.6.0 PL2
© 2002-2012 Tilted Forum Project

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360