How to find the row wise p values corresponding to each coefficient of linear regression?

1 visualización (últimos 30 días)
I have 3 matrices Y, x1, and x2 each having 3420 rows and 29 columns. I want the row wise linear regression p values of Y on x1, x2, and the interaction between x1 and x2 such that the p value term for each of them is of 3420 rows and 1 column.
I tried the following code:
nsample = 29;
nvar = 3420;
% sample data
y = rand(nsample, nvar);
x1 = rand(nsample, nvar);
x2 = rand(nsample, nvar);
xInt = x1.*x2;
intercept = ones(nsample, 1);
beta = zeros(nvar, 4);
for i = 1:nvar
desMat = [intercept, x1(:, i), x2(:, i), xInt(:, i)];
beta(i, :) = regress(y(:, i), desMat);
end
But beta only gives the linear regression coefficients. I want the p values.
How to get the p values corresponding to the linear regression coefficients of Y on x1, x2, and interaction between x1, x2 as a 3420 rows and 1 column array?
  2 comentarios
Jeff Miller
Jeff Miller el 16 de Jun. de 2022
Do you want the p value of each predictor based on its Type I, II, or III sum of squares (e.g., intro to sums of squares types)?
Abhishek Chakraborty
Abhishek Chakraborty el 17 de Jun. de 2022
I want Type I sum of squares so that the p value for each predictor is an array containing 3420 rows and 1 column.

Iniciar sesión para comentar.

Respuesta aceptada

Jeff Miller
Jeff Miller el 18 de Jun. de 2022
For Type I (i.e., sequential), you have to fit a series of three models, and you need to compute the improvement of each model over the previous one. For this application, I think it is more convenient to use the fitlm command which provides its own intercept & easier-to-use output. So, I think this will do what you are asking for:
nsample = 29;
nvar = 3420;
% sample data
y = rand(nsample, nvar);
x1 = rand(nsample, nvar);
x2 = rand(nsample, nvar);
xInt = x1.*x2;
% intercept = ones(nsample, 1); % not needed
beta = zeros(nvar, 4);
type_1_p_value_for_x1 = zeros(nvar, 1);
type_1_p_value_for_x2 = zeros(nvar, 1);
type_1_p_value_for_int = zeros(nvar, 1);
for ivar = 1:nvar
desMatx1 = [x1(:, ivar)];
mdlx1 = fitlm(desMatx1,y(:,ivar));
Fx1 = mdlx1.SSR / mdlx1.MSE;
type_1_p_value_for_x1(ivar) = 1 - fcdf(Fx1,1,mdlx1.DFE);
desMatx1x2 = [x1(:, ivar), x2(:, ivar)];
mdlx1x2 = fitlm(desMatx1x2,y(:,ivar));
Fx2givenx1 = (mdlx1x2.SSR - mdlx1.SSR) / mdlx1x2.MSE;
type_1_p_value_for_x2(ivar) = 1 - fcdf(Fx2givenx1,1,mdlx1x2.DFE);
desMatx1x2int = [x1(:, ivar), x2(:, ivar), xInt(:,ivar)];
mdlx1x2int = fitlm(desMatx1x2int,y(:,ivar));
Fintgivenx1x2 = (mdlx1x2int.SSR - mdlx1x2.SSR) / mdlx1x2int.MSE;
type_1_p_value_for_int(ivar) = 1 - fcdf(Fintgivenx1x2,1,mdlx1x2.DFE);
beta(ivar, :) = mdlx1x2int.Coefficients.Estimate;
end

Más respuestas (0)

Categorías

Más información sobre Descriptive Statistics en Help Center y File Exchange.

Productos


Versión

R2020b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by