ANCOVA - can I delete a very significant variable from a model's equation?

CGG

New Member
#1
Hi all,
I undertook a sensitivity analysis to study the contribution of explanatory variables to soil organic carbon variations. Here, the explanatory variables were the soil, climate, fertilization, tillage, stubble and rotation managements.

Since we have a linear model, a combination of R2 (coefficient of determination) of the explanatory variable alone with semi-partial R2 was an efficient way to summarize the influence of the variables on the Soil Organic Carbon. R2 of the explanatory variable i and the semi-partial R2 represent the contribution of the variable alone and the contribution of the variable with its interaction with other variables to the Soil Organic Carbon variance respectively.

General model:

E (Y| X_1,…,X_p) =α+β_1 X_1+•••+β_p X_p+ Ɛ Where Y is the response measurement Soil Organic Carbon, Xi is the explanatory variable i (soil, climate and fertilization, tillage, stubble and rotation managements), α is the intercept, the βj are the slopes or coefficients and Ɛ the errors.

My question is: the influence of the explanatory variable ‘soil’ on soil organic carbon is the biggest and very obvious so I don’t want to include it in my analyses anymore and just want to look at the influence of climate, fertilization, tillage, stubble and rotation managements. Is it correct to just simply delete it from the General model although it is a significant variable? If not, I guess another option would be to fix the value of the soil variable to a certain soil type.

Thank you for your help!:)
 

CGG

New Member
#3
Hi hlsmith,
Thank you for your time. 'soil organic carbon' is a continuous variable. The explanatory variables soil, climate, fertilization, tillage, stubble and rotation are categorical (each variable has 2 levels, except fertilisation that has 3 levels)
 

rogojel

TS Contributor
#4
hi,
omitting a significant variable from the model is not a good idea because it will increase the unaccounted for variability and make your subsequent models worse. Can't you analyse the influence of other factors in the full model?
regards
 

CGG

New Member
#5
Hi Rogojel,
Thank you very much for your reply and advise!
hi,
Can't you analyse the influence of other factors in the full model?
regards
Yes, I analyse the influence of all the explanatory variables in the full model. What I will do next is to fix the soil variable (ex soil=typeA) and look at the influence of the other explanatory variables when the soil is fixed.
 

Karabiner

TS Contributor
#6
Yes, I analyse the influence of all the explanatory variables in the full model. What I will do next is to fix the soil variable (ex soil=typeA) and look at the influence of the other explanatory variables when the soil is fixed.
This makes sense if you expect interactions between soil type and
other explanatory variables (if there aren't such interactions, then
the influence of other variables should be the same for each type
of soil). You could include the interactions directly in the model,
instead of performing separate analyses.

With kind regards

K.