ezANOVA.R

ezANOVA <-
function(
	data
	, dv
	, wid
	, within = NULL
	, between = NULL
	, observed = NULL
	, diff = NULL
	, reverse_diff = FALSE
	, detailed = FALSE
){
	to_return = ezANOVA_main(data,dv,wid,within,between,observed,diff,reverse_diff)

	########
	# Compute likelihood ratios
	########
	
	n = length(levels(to_return$data[,names(to_return$data)==wid]))	

	involves_within = laply(
		strsplit(to_return$ANOVA$Effect,':')
		, function(x){
			any(x %in% as.character(within))
		}
	)
	involves_between = laply(
		strsplit(to_return$ANOVA$Effect,':')
		, function(x){
			any(x %in% as.character(between))
		}
	)

	between_error = to_return$ANOVA$SSd[1]
	all_between_effects = sum(to_return$ANOVA$SSn[involves_between & !involves_within])
	between_null = between_error+all_between_effects
	
	within_error = sum(to_return$ANOVA$SSd[!involves_between & involves_within])
	all_within_effects = sum(to_return$ANOVA$SSn[!involves_between & involves_within])+sum(to_return$ANOVA$SSn[involves_between & involves_within])
	within_null = within_error+all_within_effects
		
	#compute the likelihood ratios for each effect
	to_return$ANOVA$model_between = NA
	to_return$ANOVA$model_within = NA
	to_return$ANOVA$comparison_between = NA
	to_return$ANOVA$comparison_within = NA
	to_return$ANOVA$LR = NA
	for(i in 2:nrow(to_return$ANOVA)){
		model_between = between_null
		model_within = within_null
		effect_split = strsplit(to_return$ANOVA$Effect[i], ':')[[1]]
		if(length(effect_split)>1){
			from_terms = terms(
				eval(
					parse(
						text = paste(
							'1~'
							, paste(
								effect_split
								, collapse = '*'
							)
						)
					)
				)
			)
			term_labels = attr(from_terms,'term.labels')
			for(this_term_label in term_labels[1:(length(term_labels)-1)]){
				if(involves_within[to_return$ANOVA$Effect==this_term_label]){
					model_within = model_within - to_return$ANOVA$SSn[to_return$ANOVA$Effect==this_term_label]
				}else{
					model_between = model_between - to_return$ANOVA$SSn[to_return$ANOVA$Effect==this_term_label]					
				}
			}
		}
		comparison_between = model_between
		comparison_within = model_within
		if(involves_within[i]){
			model_within = model_within - to_return$ANOVA$SSn[i]	
		}else{
			model_between = model_between - to_return$ANOVA$SSn[i]			
		}
		to_return$ANOVA$model_between[i] = model_between
		to_return$ANOVA$model_within[i] = model_within
		to_return$ANOVA$comparison_between[i] = comparison_between
		to_return$ANOVA$comparison_within[i] = comparison_within
		between_LR = (comparison_between/model_between)^(n/2)
		within_LR = (comparison_within/model_within)^(to_return$ANOVA$DFd[i]/2)
		to_return$ANOVA$LR[i] = between_LR*within_LR
	}	
	
	#compute and apply corrections for complexity
	aic = exp(-to_return$ANOVA$DFn)
	bic = exp(-to_return$ANOVA$DFn)^(log(n)/2)
	to_return$ANOVA$LLRa = log(aic*to_return$ANOVA$LR,base=10)
	to_return$ANOVA$LLRb = log(bic*to_return$ANOVA$LR,base=10)

	########
	# Compute effect size
	########
	if(!is.null(observed)){
		obs = rep(F,nrow(to_return$ANOVA))
		for(i in as.character(observed)){
			obs = obs | str_detect(to_return$ANOVA$Effect,i)
		}
		obs_SSn1 = sum(to_return$ANOVA$SSn*obs)
		obs_SSn2 = to_return$ANOVA$SSn*obs
	}else{
		obs_SSn1 = 0
		obs_SSn2 = 0
	}
	to_return$ANOVA$ges = to_return$ANOVA$SSn/(to_return$ANOVA$SSn+sum(unique(to_return$ANOVA$SSd))+obs_SSn1-obs_SSn2)
	
	########
	# Final clean-up
	########

	#remove the data from to_return
	temp = names(to_return)
	temp = temp[temp!='data']
	to_return = to_return[names(to_return)!='data']
	names(to_return) = temp

	#if necessary, remove extra columns and the Intercept row from the anova
	if(!detailed){
		to_return$ANOVA = to_return$ANOVA[,!(names(to_return$ANOVA) %in% c('SSn','SSd','between_LR','within_LR','LR','k','aic','bic'))]
		to_return$ANOVA = to_return$ANOVA[2:nrow(to_return$ANOVA),]
	}

	#all done!
	return(to_return)
}