R help - Aug 2010 - Kolmogorov-Smirnov test, which one to use?

Hi,

I have two sets of data, an observed data and generated data.
The generated data is obtained from the model where the parameters is estimated 
from the observed data.

So I'm not sure which to use either
one-sample test 
ks.test(x+2, "pgamma", 3, 2) # two-sided, exact

or

two-sample test 
ks.test(x, x2, alternative="l")

If I use the one-sample test I need to specified the model which I don't
have in
my case.

Actually I use the two-sample test and when I compare with what I got from using
Chi-square test the result is too different.

Data:

      obs_data  pre_gam
 [1,]       93  25.6770
 [2,]      115 127.9095
 [3,]      125 151.6845
 [4,]      120 146.9295
 [5,]      106 107.9385
 [6,]      101 107.4630
 [7,]       75  86.5410
 [8,]       58  55.6335
 [9,]       46  43.7460
[10,]       38  32.8095
[11,]       31  16.1670
[12,]       17  18.5445
[13,]       10   9.0345
[14,]       16  20.9220

Results:
> chisq.test(obs_data, p = pre_gam, rescale.p = TRUE)
        Chi-squared test for given probabilities
data:  obs_data 
X-squared = 205.4477, df = 13, p-value < 2.2e-16
> ks.test(obs_data,pre_gam)
        Two-sample Kolmogorov-Smirnov test
data:  obs_data and pre_gam 
D = 0.2143, p-value = 0.9205
alternative hypothesis: two-sided


Am I doing the right thing?

Thank you so much for your help.


      
	[[alternative HTML version deleted]]

It is not clear what question you are trying to answer.  Perhaps if you can give
us an explanation of your overall goal then we can be more helpful.

-- 
Gregory (Greg) L. Snow Ph.D.
Statistical Data Center
Intermountain Healthcare
greg.snow at imail.org
801.408.8111

> -----Original Message-----
> From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-
> project.org] On Behalf Of Roslina Zakaria
> Sent: Wednesday, August 04, 2010 8:34 PM
> To: r-help at r-project.org
> Subject: [R] Kolmogorov-Smirnov test, which one to use?
> 
> Hi,
> 
> I have two sets of data, an observed data and generated data.
> The generated data is obtained from the model where the parameters is
> estimated
> from the observed data.
> 
> So I'm not sure which to use either
> one-sample test
> ks.test(x+2, "pgamma", 3, 2) # two-sided, exact
> 
> or
> 
> two-sample test
> ks.test(x, x2, alternative="l")
> 
> If I use the one-sample test I need to specified the model which I
> don't have in
> my case.
> 
> Actually I use the two-sample test and when I compare with what I got
> from using
> Chi-square test the result is too different.
> 
> Data:
> 
> ????? obs_data? pre_gam
> ?[1,]?????? 93? 25.6770
> ?[2,]????? 115 127.9095
> ?[3,]????? 125 151.6845
> ?[4,]????? 120 146.9295
> ?[5,]????? 106 107.9385
> ?[6,]????? 101 107.4630
> ?[7,]?????? 75? 86.5410
> ?[8,]?????? 58? 55.6335
> ?[9,]?????? 46? 43.7460
> [10,]?????? 38? 32.8095
> [11,]?????? 31? 16.1670
> [12,]?????? 17? 18.5445
> [13,]?????? 10?? 9.0345
> [14,]?????? 16? 20.9220
> 
> Results:
> > chisq.test(obs_data, p = pre_gam, rescale.p = TRUE)
> ??????? Chi-squared test for given probabilities
> data:? obs_data
> X-squared = 205.4477, df = 13, p-value < 2.2e-16
> 
> > ks.test(obs_data,pre_gam)
> ??????? Two-sample Kolmogorov-Smirnov test
> data:? obs_data and pre_gam
> D = 0.2143, p-value = 0.9205
> alternative hypothesis: two-sided
> 
> 
> Am I doing the right thing?
> 
> Thank you so much for your help.
> 
> 
> 
> 	[[alternative HTML version deleted]]