Nested design model and

cross-nested
1. Nested design models
Nested or hierarchical models are used in some
situations where all levels of a factor cannot be combined with
all levels of another, that is, not all possible ones can be determined
treatments that appear when crossing the factors.

Example 1.
Let's assume that in a vocational training center, the study is about
percentage of Approved students in a subject, in the morning and afternoon groups.
In the morning, three people teach the subject, and in the afternoon, two.
class person to three groups and these are supposed to be replicas (they are not
source of variation).
Thus,

As can be seen, people are 'tied' to the shift they teach.

the classes and differences could be observed when taking measurements at
the replicas in this case group of students then it is said that factor B
is nested in factor A, namely B⊂ A.

Example 2
Consider a company that buys its raw material from three suppliers.
different. The company wants to determine if the purity of the raw material of
Each supplier is the same. There are four batches of raw material available from
each supplier, and three determinations of purity will be made for each batch.
At first glance, one might wonder why it is not a factorial experiment. If
if it were a factorial experiment, then batch 1 would always refer to the same
batch, batch 2 would always refer to the same batch, etc.
Clearly, this is not the case, since the batches from each supplier are
unique to that particular supplier. That is, lot 1 from supplier 1 does not
it is related to lot 1 of any other supplier. As seen in the
next graph.

2. Mathematical model

It is said that factor B is nested within another factor A (or that its levels are
nested in those of A) when each level of factor B appears associated with a
unique level of the factor

It is denoted as B⊂ A.

The model is expressed as

Where
i = 1, ... , a
j = 1, ..., b
k = 1, ... , n
Y
For each:
It is observed that βj(i) represents the additional average effect of the j-th level.
nested at level i.
On the other hand, b is the number of nested levels at each level i, so that
the total number of levels of B is a · b and the sum of the effects of factor B
within each level of A is 0.

Least squares estimators

It has to be

Thus,

For each fixed i

For each fixed i and fixed j

In this way

The total number of observations is a·b·n and the total number of parameters is a
estimating is 1+(a−1)+a(b−1) = ab, then the total number of degrees of freedom is
ab - ab = ab(n - 1). In this way, the estimate of the variance is
ANOVA Table
If we consider the total sum of squares

By adding and subtracting the terms ±¯yi··±¯yij· one obtains

So

SCT = SCA + SCB(A) + SCE

What position does it hold in terms of totals?

For each fixed level i, there is

The hypothesis contrasts that are performed are:

In this case

Thus, H0 is rejected at level α if

The other hypothesis being contrasted is,∀i = 1, . . . , a

In this case,

Thus, H0 is rejected at level α if

Fa(b−1), ab(n−1), α

In this case, the hypothesis is contrasted that all levels of the factor
nested B are equal within the factor A where they are nested.
However, if it is found that they are different on a global scale, it is interesting.
compare, next, if the levels of factor B nested in A are equal
among themselves, within each level i (of A) in which they are nested.
Thus, for each fixed level of i, where i = 1, ..., a, it is contrasted whether the levels
whether the nested factor levels are equal or not within each of the levels of the factor
In which they are individually nested
In this case

Thus, H0 is rejected at level α if

Example.1

A geologist studies the trace radioactive content of five different types.

of soil. To do this, it collects four samples of content in substances
radioactive in four different locations that are situated on each type
from the ground. The following data is obtained:
It is obtained that
F (a−1), ab(n−1), α = F4, 60, 001 = 2, 04

Therefore it is accepted
H0 : α1 = · · · = αa = 0

Thus, there are no significant differences between the lands at the α = 0.1 level.
On the other hand,

Fa(b−1), ab(n−1), α = F15,60,001 = 1,6

Then it is rejected∀i = 1,..., to the hypothesis H0: β1(i)= ... = βb(i)= 0 at level α =
0,1.
We study the individual contrasts by level.
It is calculated for each i

In this way,

The ANOVA table is as follows

How

There are significant differences in levels B(2) and C(3), that is, regarding
Type B and C lands have different levels of substance in the localities.
radioactive.

Nested models of random effects and effects

mixed

Random effects model

The mathematical model is

Being

Where

For all i,

Where all the random variables are independent.

Thus,
They are mainly independent.
The expectations of the mean squares are:

The estimates of the variance components are:

The contrasts of hypotheses are:

Mixed effects model B⊂ A, (A fixed, B random)

The model is

Being

Where

For every i,
Where all the random variables are independent.
Thus,

They are mainly independent.

The hopes of the mean squares are:

The estimates of the variance components are:

Cross-nested design models

It is said that two factors are completely crossed when they appear
all possible combinations of the levels of each factor, as is the
case of a two-factor design.

It is said that two factors are crossed when neither of them is nested.
in the other, that is, neither A⊂ B is B⊂ A.
For example, in the previous example of shifts and teaching staff, it may be that
some of these teach in both shifts at the same time, resulting in having two
crossed factors although not completely crossed.
Cross-nested designs are characterized by having both factors
crossed like nested. There is no single mathematical model, since
it depends on the arrangement of the factors in the design.
Model 1.
This is about studying the assembly time of a series of watchmaking pieces.
that must be done by hand. Three different positions are considered for assembly
the pieces and four different sizes of them. The assembly is carried out by
a series of people, so that two different people are occupied for
each assembly with each size and position.

It is thus taken as factors,

It is observed that all levels of the position factor intersect with all the
size factor levels and that the individual factor has different levels for
each of the crossings, that is, different people work in each case. The
the scheme is P∗ T and I⊂ (P∗T).
Mathematical model. The model must include:
The main effects of P and T.
The interactions between P and T.
The effects of each level k of the nested factor I in the combination of (i,j).

Where
i = 1, . . . , a
j = 1, ..., b
k = 1, . . . , c
l = 1, . . . , n
Subject to restrictions

∀(ij) fixed

Independents from each other.

The hopes of the squares means son

The estimates of the variance components are:

Rule to determine the means of squares

expected (by Hesse)
In principle, it is necessary to distinguish between fixed factors and random ones. The
Interactions between fixed and random factors are considered random.
Each random factor is assigned its variance component σ2.
fixed factor is assigned its fixed effect represented by the sum of squares of
the parameters associated with that factor among its degrees of freedom, e.g. if A is
a fixed factor is associated
The following table is constructed:

1. The term of the model error εij...mse is represented as ε(ij...)m where m is the
index of the replications, that is, fixed ij . . . m replicas are considered
random.

2. The subscripts of each term are divided into three classes:

— assets: they are in the term and are not in parentheses.
— liabilities: they are in the term but are in parentheses.
— absent: They are not in the term although they belong to the model.

3. In each row, a 1 is written if one of the passive subscripts is present (they are also
in parentheses those corresponding to the nested ones) of the row component
matches the subscript of the column.

4. If any subscript of the row matches the subscript of the column, it

write:

A 1 is a random factor
— A 0 if it is a fixed factor
In the remaining gaps, the number of levels in the column is written.
corresponding.
The following table is as follows:

5. To obtain the expected value of the mean of squares of any

component of the model, it is done:
a) All columns headed by the active subscripts of that are covered.
component.
b) The numbers of the rows that have at least the same are multiplied
subindices that the component, multiplying them in turn by the fixed factor (sum
of squares) or the random factor (variance) obtained.
c) The sum of these products is the expected value of the mean of squares.
In the example, it would be:

Therefore, the contrasts of hypotheses are:

In any case, hypothesis contrasts cannot always be constructed for
any model. One possible solution is to assume that some interactions
are null.

Example. Suppose a bifactorial model with random effects:

Therefore, the contrasts of hypotheses are:

Example. Let's assume a bifactorial model with fixed effects:

Therefore, the hypothesis contrasts are:

Example. Let's suppose a nested model with two factors: B⊂ Where B is

random and A is fixed:

Where

For all i,
Where all the random variables are independent.

The checked tights are:

Therefore, the hypothesis contrasts are:

Example. Let's assume a completely randomized block model with a

random factor and a random block

Being

Where

Where all the random variables are independent.

In this case, there are no random replicas and there is only one fixed element (k = 1).
In this way, k is not a replica, which is why it is set.
The square stockings are:

Therefore, the hypothesis contrast is:

Example. Let's assume a complete randomized block model with a

random factor and a fixed block:

Where

Where all the random variables are independent.

In this case, there are no random replicas and there is only one fixed element (k = 1).
Thus, k is not a replica so it is put εij.
The square socks are: