On Tuesday, October 14, 2014 6:57:57 AM UTC-5, David Gonzales wrote:

The thing with julia is that most of the language is written in julia, so 
> getting answers means just reading more julia code. So in GLM/src/lm.jl 
> there is a function coeftable(..) that generates the above table. Taking 
> the calculation from there, you get:
>
> [ccdf(FDist(1,df_residual(lm1.model)),abs2(fval)) for fval in 
> coef(lm1)./stderr(lm1)]
>
> which gives the Pr(>|t|) column:
>  0.551595  
>  3.40919e-7
>
> And as for how it is calculated - the formula shows it uses the F 
> distribution.
>

That piece of code may be a little too terse.  The column name "Pr(>|t|) " 
is intended to be read "the probability of exceeding the absolute value of 
the observed t-statistic" which is the p-value for a test of the 
coefficient in question being zero versus not equal to zero.  Implicit in 
the test is the distribution of that ratio as a Student's T distribution 
with n-p degrees of freedom where n is the total number of observations and 
p is the number of coefficients.

It happens that the square of a Student's T distribution on n-p degrees of 
freedom is an F distribution with 1 and n-p degrees of freedom and it is 
easier to evaluate the probability of the F distribution exceeding the 
square of the t-statistic, which is what is done here.

>
> On Tuesday, October 14, 2014 12:21:17 PM UTC+3, [email protected] wrote:
>>
>> In below example of GLM, I want to get the Pr(>|t|) value 3.4e-7. How can I 
>> get it?
>> Also, how this p-value be calculated? By F test or by Chisq test? I can 
>> choose the test type in R but I can not choose in Julia.
>>
>> julia> using GLM, RDatasets
>> julia> form = dataset("datasets","Formaldehyde")6x2 
>> DataFrame|-------|------|--------|| Row # | Carb | OptDen || 1     | 0.1  | 
>> 0.086  || 2     | 0.3  | 0.269  || 3     | 0.5  | 0.446  || 4     | 0.6  | 
>> 0.538  || 5     | 0.7  | 0.626  || 6     | 0.9  | 0.782  |
>> julia> lm1 = fit(LinearModel, OptDen ~ Carb, form)Formula: OptDen ~ Carb
>> Coefficients:
>>                Estimate  Std.Error  t value Pr(>|t|)(Intercept)  0.00508571 
>> 0.00783368 0.649211   0.5516Carb           0.876286  0.0135345  64.7444   
>> 3.4e-7
>>
>>

Reply via email to