Correlate

A somewhat clunky solution:
for(i in colnames(dat)){
  print(cor.test(dat[,i], dat$x1, method = "pearson", use = "complete.obs")$estimate)
  print(cor.test(dat[,i], dat$x1, method = "pearson", use = "complete.obs")$p.value)
}

Rather than printing you could set up an array or list to save the results.


Tim

-----Original Message-----
From: R-help <r-help-bounces at r-project.org> On Behalf Of Val
Sent: Monday, August 22, 2022 11:09 AM
To: r-help at R-project.org (r-help at r-project.org) <r-help at r-project.org>
Subject: [R] Correlate

[External Email]

Hi all,

I have a data set with  ~250  variables(columns).  I want to calculate the correlation of  one variable with the rest of the other variables and also want  the p-values  for each correlation.  Please see the sample data and my attempt.  I  have got the correlation but unable to get the p-values

dat <- read.table(text="x1 x2 x3 x4
           1.68 -0.96 -1.25  0.61
          -0.06  0.41  0.06 -0.96
              .    0.08  1.14  1.42
           0.80 -0.67  0.53 -0.68
           0.23 -0.97 -1.18 -0.78
          -1.03  1.11 -0.61    .
           2.15     .    0.02  0.66
           0.35 -0.37 -0.26  0.39
          -0.66  0.89   .    -1.49
           0.11  1.52  0.73  -1.03",header=TRUE)

#change all to numeric
    dat[] <- lapply(dat, function(x) as.numeric(as.character(x)))

    data_cor <- cor(dat[ , colnames(dat) != "x1"],  dat$x1, method = "pearson", use = "complete.obs")

Result
              [,1]
x2 -0.5845835
x3 -0.4664220
x4  0.7202837

How do I get the p-values ?

Thank you,

A somewhat clunky solution:
for(i in colnames(dat)){
  print(cor.test(dat[,i], dat$x1, method = "pearson", use = "complete.obs")$estimate)
  print(cor.test(dat[,i], dat$x1, method = "pearson", use = 
"complete.obs")$p.value) }

Rather than printing you could set up an array or list to save the results.


Tim

-----Original Message-----
From: R-help <r-help-bounces at r-project.org> On Behalf Of Val
Sent: Monday, August 22, 2022 11:09 AM
To: r-help at R-project.org (r-help at r-project.org) <r-help at r-project.org>
Subject: [R] Correlate

[External Email]

Hi all,

I have a data set with  ~250  variables(columns).  I want to calculate 
the correlation of  one variable with the rest of the other variables 
and also want  the p-values  for each correlation.  Please see the 
sample data and my attempt.  I  have got the correlation but unable to 
get the p-values

dat <- read.table(text="x1 x2 x3 x4
           1.68 -0.96 -1.25  0.61
          -0.06  0.41  0.06 -0.96
              .    0.08  1.14  1.42
           0.80 -0.67  0.53 -0.68
           0.23 -0.97 -1.18 -0.78
          -1.03  1.11 -0.61    .
           2.15     .    0.02  0.66
           0.35 -0.37 -0.26  0.39
          -0.66  0.89   .    -1.49
           0.11  1.52  0.73  -1.03",header=TRUE)

#change all to numeric
    dat[] <- lapply(dat, function(x) as.numeric(as.character(x)))

    data_cor <- cor(dat[ , colnames(dat) != "x1"],  dat$x1, method = 
"pearson", use = "complete.obs")

Result
              [,1]
x2 -0.5845835
x3 -0.4664220
x4  0.7202837

How do I get the p-values ?

Thank you,

I (maybe) agree, but I would go further than that. There are assumptions associated with the test that are missing. It is not clear that the relationships are all linear. Regardless of a "significant outcome" all of the relationships need to be explored in more detail than what is provided in the correlation test.

Multiplicity adjustment as in : https://www.sciencedirect.com/science/article/pii/S0197245600001069 is not an issue that I can see in these data from the information provided. At least not in the same sense as used in the link.

My first guess at the meaning of "multiplicity adjustment" was closer to the experimentwise error rate in a multiple comparison procedure. https://dictionary.apa.org/experiment-wise-error-rateEssentially, the type 1 error rate is inflated the more test you do and if you perform enough tests you find significant outcomes by chance alone. There is great significance in the Redskins rule: https://en.wikipedia.org/wiki/Redskins_Rule.

A simple solution is to apply a Bonferroni correction where alpha is divided by the number of comparisons. If there are 250, then 0.05/250 = 0.0002. Another approach is to try to discuss the outcomes in a way that makes sense. What is the connection between a football team's last home game an the election result that would enable me to take another team and apply their last home game result to the outcome of a different election?

Another complication is if variables x2 through x250 are themselves correlated. Not enough information was provided in the problem to know if this is an issue, but 250 orthogonal variables in a real dataset would be a bit unusual considering the experimentwise error rate previously mentioned.

Large datasets can be very messy.


Tim

-----Original Message-----
From: Bert Gunter <bgunter.4567 at gmail.com>
Sent: Monday, August 22, 2022 12:07 PM
To: Ebert,Timothy Aaron <tebert at ufl.edu>
Cc: Val <valkremk at gmail.com>; r-help at R-project.org (r-help at r-project.org) <r-help at r-project.org>
Subject: Re: [R] Correlate

[External Email]

... But of course the p-values are essentially meaningless without some sort of multiplicity adjustment.
(search on "multiplicity adjustment" for details). :-(

-- Bert


On Mon, Aug 22, 2022 at 8:59 AM Ebert,Timothy Aaron <tebert at ufl.edu> wrote:

A somewhat clunky solution:
for(i in colnames(dat)){
  print(cor.test(dat[,i], dat$x1, method = "pearson", use = "complete.obs")$estimate)
  print(cor.test(dat[,i], dat$x1, method = "pearson", use =
"complete.obs")$p.value) }

Rather than printing you could set up an array or list to save the results.


Tim

-----Original Message-----
From: R-help <r-help-bounces at r-project.org> On Behalf Of Val
Sent: Monday, August 22, 2022 11:09 AM
To: r-help at R-project.org (r-help at r-project.org) <r-help at r-project.org>
Subject: [R] Correlate

[External Email]

Hi all,

I have a data set with  ~250  variables(columns).  I want to calculate
the correlation of  one variable with the rest of the other variables
and also want  the p-values  for each correlation.  Please see the
sample data and my attempt.  I  have got the correlation but unable to
get the p-values

dat <- read.table(text="x1 x2 x3 x4
           1.68 -0.96 -1.25  0.61
          -0.06  0.41  0.06 -0.96
              .    0.08  1.14  1.42
           0.80 -0.67  0.53 -0.68
           0.23 -0.97 -1.18 -0.78
          -1.03  1.11 -0.61    .
           2.15     .    0.02  0.66
           0.35 -0.37 -0.26  0.39
          -0.66  0.89   .    -1.49
           0.11  1.52  0.73  -1.03",header=TRUE)

#change all to numeric
    dat[] <- lapply(dat, function(x) as.numeric(as.character(x)))

    data_cor <- cor(dat[ , colnames(dat) != "x1"],  dat$x1, method =
"pearson", use = "complete.obs")

Result
              [,1]
x2 -0.5845835
x3 -0.4664220
x4  0.7202837

How do I get the p-values ?

Thank you,