I noticed that the name of an interaction term between two variables is "variable1.x.variable2".
I understand the meaning of multiplication, but I was wondering whether you could change it to "variable1:variable2", which is closer to what lm() and glm() show in the summary.
Another solution would be to make remstats return the array of statistics without interactions or transformation terms, e.g., I(x^2), and remstimate will handle that step internally. This workflow will be closer to what lm() and glm() do, which will result in a smaller output array, and less space in memory required.
Of course, you could also change the meaning of the input formula of remstats and make it not about the modelling but about computing a set of statistics that then will be modelled in remstimate by specifying a linear predictor of the model where there you include interactions and transformations. This should also reduce the number of times an array of statistics is calculated after different model specifications because given an initial set of statistics the user can specify different models and run different estimations in remstimate starting from the same array of statistics.
Hi Marlyne,
I noticed that the name of an interaction term between two variables is
"variable1.x.variable2".I understand the meaning of multiplication, but I was wondering whether you could change it to
"variable1:variable2", which is closer to whatlm()andglm()show in the summary.Another solution would be to make
remstatsreturn the array of statistics without interactions or transformation terms, e.g., I(x^2), andremstimatewill handle that step internally. This workflow will be closer to whatlm()andglm()do, which will result in a smaller output array, and less space in memory required.Of course, you could also change the meaning of the input formula of
remstatsand make it not about the modelling but about computing a set of statistics that then will be modelled in remstimate by specifying a linear predictor of the model where there you include interactions and transformations. This should also reduce the number of times an array of statistics is calculated after different model specifications because given an initial set of statistics the user can specify different models and run different estimations in remstimate starting from the same array of statistics.What do you think?
Cheers, Giuseppe