[R] Linear Regression with 2 grouping variables

[Michael Dewey]

At 02:15 23/08/2011, Nathan Miller wrote:
>Hi all,

See comment in-line

>I have a data set that looks a bit like this.
>
>feed1
>       RFU Site Vial Time       lnRFU
>1   44448    1    1   10  10.702075
>2   47521    1    1   20  10.768927
>3   42905    1    1   30  10.66674
>4   46867    1    1   40  10.755069
>5   42995    1    1   50  10.668839
>6   43074    1    1   60  10.670675
>7   41195    1    1   70  10.626072
>8   47090    1    2   10  10.759816
>9   48100    1    2   20  10.781037
>10  43215    1    2   30  10.673943
>11  39656    1    2   40  10.587998
>12  38799    1    2   50  10.566150
>13  38424    1    2   60 10.556438
>14  35240    1    2   70  10.469937
>15  46427    1    3   10  10.745636
>16  46418    1    3   20  10.745443
>17  42095    1    3   30  10.647684
>......
>There are 5 columns of data, three levels of "Site", 10 "Vials" per site,
>and measurements were taken at 10 min intervals from 10-70.. I am primarily
>interested in the relationship between "Time" and "lnRFU" to calculate the
>rate at which lnRFU declines over time. I have a nice plot using a ggplot2
>code that looks like this
>
>p<-ggplot(data=feed1,aes(x=Time,y=lnRFU))
>p+geom_point(size=4)+facet_grid(Site~Vial)+geom_smooth(method="lm")
>
>The graph is useful to visualize the changes over time and grouped by both
>Site and Vial, but I also need the slopes of the linear regressions for each
>Vial, within a Site. This is where I run into a problem. I want to run a
>linear regression of lnRFU as a function of Time grouped by both Site and
>Vial. Its easy to visualize this comparison in ggplot using facet_grid(),
>but I'm not sure how to do a similar comparison/analysis within lm()
>
>I imagine something like
>
>fit<-lm(lnRFU~Time | Vial * Site, data=feed1)

I think you will find lmList from nlme helpful here, try
library(nlme)
fitlist <- lmList(lnRFU~Time | Vial * Site, data=feed1)

Untested, but should be OK. There are a number of helper functions in 
nlme which operate on lmList objects to give handy plots and more.

>  in which I group by both Vial and Site, but obviously this code doesn't
>work. Does anyone have an idea for how to do a linear regression with two
>grouping variables? Do I have to go back and combine Vial and Site into a
>single grouping variable or can I leave the dataframe the way it is? I'm
>trying to imagine a means of accomplishing the same type of thing that
>facet_grid does when it allows you to plot the data as a function of two
>"grouping" variables.
>
>Thanks for you time. I greatly appreciate it.
>
>Nate Miller
>
>         [[alternative HTML version deleted]]

Michael Dewey
info at aghmed.fsnet.co.uk
http://www.aghmed.fsnet.co.uk/home.html