"Vectorizing" this for-loop in R? (suppressing interaction main effects in lm) -

when interactions specified in lm, r includes main effects default, no option suppress them. appropriate , convenient, there instances (within estimators, ratio lhs variables, among others) isn't appropriate.

i've got code fits log-transformed variable response variable, independently within subsets of data.

here silly yet reproducible example:

id = as.factor(c(1,2,2,3,3,3,4,4,4,4,5,5,5,5,6,7,7,8,8,8,9,9,9,9,10)) x = rexp(length(id)) y = rnorm(length(id)) logx = log(x) data = data.frame(id,y,logx)  (i in data$id){     sub = subset(data, id==i)   #this splits data id     m = lm(y~logx-1,data=sub)   #this gives me linear (log) fit 1 of id's     sub$x.tilde = log(1+3)*m$coef   #this linearizes , gives me expected value x=3     data$x.tilde[data$id==i] = sub$x.tilde #this puts main dataset     data$tildecoeff[data$id==i] = m$coef #this saves coefficient (i use elsewhere plotting)     } 

i want fit model following:

y = b(x*id) +e

with no intercept , no main effect of id. can see loop, i'm interested in expectation of y when x=3, constrained fit through origin (because y (logged) ratio of y[x=something]/y[x=0].

but if specify

m = lm(y~x*as.factor(id)-1) 

there no means of suppressing main effects of id. need run loop several hundred times in iterative algorithm, , loop far slow.

the other upside of de-looping code it'll more convenient prediction intervals.

(please, don't need pious comments how leaving out main effects , intercepts improper -- is, can promise isn't in instance).

thanks in advance ideas!

i think want

m <- lm(y ~ 0 + logx : as.factor(id)) 

see r-intro '11.1 defining statistical models; formulae'