orgmode and anova

orgmode and anova

[Uwe Brauer]

Hello

Using Kubuntu I just installed R and the following code works nicely

#+tblname: delsee
| airmass | zenith_seeing | delivered_seeing |
|---------+---------------+------------------|
|     1.3 |          0.95 |        1.1119612 |
|     1.3 |           1.0 |        1.1704854 |
|     1.3 |           1.1 |        1.2875340 |
|     1.3 |           1.2 |        1.4045825 |
#+TBLFM: $3=$2*($1**0.6)


#+begin_src R :results output :var delsee=delsee 
summary(delsee)
#+end_src


Does somebody know whether I could do an ANOVA, comparing these columns
(which does not make much sense, but this is not the point.

Any help is strongly appreciated.

thanks

Uwe Brauer 

Re: orgmode and anova

[Kyle Andrews]

[-- Attachment #1: Type: text/plain, Size: 1526 bytes --]

If you wanted to fit a linear regression model with R, you could do so with
the lm function. Calling anova on the output of the regression would give
you a regression anova table.

model <- lm(delivered_seeing ~ zeenith_seeing, data = delsee)
anova(model)

You would need non-zero residuals for that to be useful of course.

Otherwise, you need to stack your *_seeing columns into one column with
another column saying which kind of seeing it was and then:

model.aov <- aov(seeing ~ factor(kind), data = delsee2)

You could do the stacking in a number of ways. My favorite is to use the
gather function in the tidyr package.

aov is just a wrapper around lm, so just take the same approach as before
to get the ANOVA table. Hope that helps.

On Thu, Apr 21, 2016, 15:10 Uwe Brauer <oub@mat.ucm.es> wrote:

> Hello
>
> Using Kubuntu I just installed R and the following code works nicely
>
> #+tblname: delsee
> | airmass | zenith_seeing | delivered_seeing |
> |---------+---------------+------------------|
> |     1.3 |          0.95 |        1.1119612 |
> |     1.3 |           1.0 |        1.1704854 |
> |     1.3 |           1.1 |        1.2875340 |
> |     1.3 |           1.2 |        1.4045825 |
> #+TBLFM: $3=$2*($1**0.6)
>
>
> #+begin_src R :results output :var delsee=delsee
> summary(delsee)
> #+end_src
>
>
> Does somebody know whether I could do an ANOVA, comparing these columns
> (which does not make much sense, but this is not the point.
>
> Any help is strongly appreciated.
>
> thanks
>
> Uwe Brauer
>
>
>

[-- Attachment #2: Type: text/html, Size: 2027 bytes --]

Re: orgmode and anova

[Uwe Brauer]

>>> "Kyle" == Kyle Andrews <kyle.c.andrews@gmail.com> writes:

   > If you wanted to fit a linear regression model with R, you could do
   > so with the lm function. Calling anova on the output of the
   > regression would give you a regression anova table.

   > model <- lm(delivered_seeing ~ zeenith_seeing, data = delsee)
   > anova(model)

   > You would need non-zero residuals for that to be useful of course.

   > Otherwise, you need to stack your *_seeing columns into one column
   > with another column saying which kind of seeing it was and then:

   > model.aov <- aov(seeing ~ factor(kind), data = delsee2)

   > You could do the stacking in a number of ways. My favorite is to use
   > the gather function in the tidyr package.

   > aov is just a wrapper around lm, so just take the same approach as
   > before to get the ANOVA table. Hope that helps.


Yes to a certain extend. I was able to perform an anova just comparing
the means of three samples. I did this in the following way:

First the data in table form


#+tblname: myanova
| C1   | C2 | C3 |
|------+----+----|
|   19 | 20 | 16 |
|   22 | 21 | 15 |
|   20 | 33 | 18 |
|   18 | 27 | 26 |
|   25 | 40 | 17 |

Then the following command did precisely  what I wanted
#+begin_src R :results output org wrap :exports results
a = c(19,22,20,18,25)
b = c(20,21,33,27,40)
c = c(16,15,18,26,17)
dati = c(a, b, c)
groups = factor(rep(letters[1:3], each = 5))
fit = lm(formula = dati ~ groups)
anova(fit)
#+end_src

I actually do not really understand precisely the meaning of
groups = factor(rep(letters[1:3], each = 5))
but I also do not care much

Now executing the code, resulted in  


#+RESULTS:
#+BEGIN_SRC org
Analysis of Variance Table

Response: dati
          Df Sum Sq Mean Sq F value  Pr(>F)  
groups     2 260.93 130.467  4.0061 0.04648 *
Residuals 12 390.80  32.567                  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 
#+END_SRC

The only thing I was interested in was the F and the p valor.


However now comes the question. I don't want to write the data in vector
form as in
a = c(19,22,20,18,25)

I want to extract them from the table:
#+tblname: myanova
| C1 1 | C2 | C3 |
|   19 | 20 | 16 |
|   22 | 21 | 15 |
|   20 | 33 | 18 |
|   18 | 27 | 26 |
|   25 | 40 | 17 |

(Rationale: I have a huge table with 300 entries and I don't want to
transform each colum in question into a vector)

So this is more a orgmode than a R question: is there a way to extract
from a given table its columns and perform via R an anova??

(I could ask the same for matlab, for which I have to transform the
table into a matlab matrix, which again I want to avoid.)

By hope is there is a sort of generalisation of my first example:


#+tblname: myan1
| growth | sugar |
|--------+-------|
|     75 | 45    |
|     72 | 44    |
|     73 | 89    |
|     61 | 40    |
|     67 | 23    |
|     64 | 33    |
|     62 | 22    |
|     63 | 22    |

#+begin_src R :results output :var myan1=myan1 
summary(myan1)
#+end_src

The command summary directly extracts the information from that table. I
hope that something similar is possible for the anova.

thanks

Uwe Brauer