In CIP-RIU/hidap: High Interactive Data Analysis Platform for Plant breeding

r i = {{i}}

# Check design

lc <- ck.f(traits[i], factors, rep, dfr)

# Fit a model for assumptions plots

expr <- paste(traits[i], '~', factors[1])
for (j in 2:lc$nf)
  expr <- paste(expr, '*', factors[j])

if (design == "crd")
  ff <- as.formula(expr)

if (design == "rcbd") {
  expr <- paste(expr, '+', rep)
  ff <- as.formula(expr)
}

model <- aov(ff, dfr)

# Estimate missing values

trait.est <- paste0(traits[i], ".est")

if (lc$nmis > 0) {
  dfr[, trait.est] <- mve.f(traits[i], factors, rep, design, dfr, maxp)[, trait.est]
} else {
  dfr[, trait.est] <- dfr[, traits[i]]
}

# Get anova table with estimated missing values

at <- suppressWarnings(aov.f(traits[i], factors, rep, design, dfr, maxp))

# CV

rr <- dim(at)[1]
cv <- (at[rr, 3])^0.5 / mean(dfr[, trait.est]) * 100

{{i+1}}. Analysis for trait `r traits[i]`

r if (lc$nmis == 0) {"There are no missing values for this trait; the design is balanced."}

r if (lc$nmis > 0) paste0("There are some missing values (", format(lc$pmis * 100, digits = 3), "%) and they have been estimated for the descriptive statistics and ANOVA.")

{{i+1}}.1. Descriptive statistics

{{i+1}}.1.1. Means by individual factor levels

for (j in 1:lc$nf)
  print(tapply(dfr[, trait.est], dfr[, factors[j]], mean))

{{i+1}}.1.2. Means by factor levels combinations

# Create expression for list of factors

lf.expr <- 'list(dfr[, factors[1]]'

for (j in 2:lc$nf)
  lf.expr <- paste0(lf.expr, ', dfr[, factors[', j, ']]')

lf.expr <- paste0(lf.expr, ')')

# Compute means over replications

tapply(dfr[, trait.est], eval(parse(text = lf.expr)), mean)

{{i+1}}.2. ANOVA

{{i+1}}.2.1. Checking assumptions

As it was stated in section 1, it is supposed that the error has a normal distribution with the same variance for all the combinations among the levels of the factors. The following plots help to evaluate this assumptions:

par(mfrow = c(1, 2))
suppressWarnings(plot(model, which = 1))
suppressWarnings(plot(model, which = 2))

Funnel shapes for the first plot may suggest heterogeneity of variances while departures from the theoretical normal line are symptoms of lack of normality.