Fixed, Random and Mixed effect models
A statistical model is usually a linear relation of the effects of the different levels of a
number of factors involved in an experiment with one error term .The effects of any factor
can be fixed, random or mixed .
In fixed effect models, the main objectives are to estimate the effects find a measure of the
variability among the effects of each of the factors and finally variability among the error
terms. On the other hand the main emphasis is on estimating the variability among the
effects of different factors(variance component) These are also referred as model I and
model II. In model I or fixed effect model the levels of each of the factor included in the
model does not represent a random sample from the population of the factor levels. Every
time an experiment is conducted, same a levels of factor A and same b levels of factor b
will be used in two factor experiment. The conclusion drawn from the analysis of the
experiment will be valid only to a levels of factor A and b levels of factor B. On the other
hand in random effect model the levels of each of the factor denote a random sample from
a large population of the factor levels. Thus in two factor model, a levels of factor A denotes
a random sample from a large population of levels of A and b levels of factor B represents a
random sample from large population of levels of B. Inferences drawn from the analysis of
variance will be valid to whole population of factor levels.
In case of mixed effect model some of the factors show fixed effect whereas some other
show random effect. In other words in two factor model the levels of factor A may be
considered as fixed where as the levels of factor B may represent a random sample from
large population of levels of factor B. Interaction effect of factor A & B is also random.
The linear model considered in all the three types of models involving two factors is same
and given as below
Yijk = µ + αi + βj + (αβ)ij + €ijk i=1,2….. a and j= 1,2…….b are a levels of factor A and b levels
of factor B respectively α i denotes the effect of ith level of factor A and β j denotes the
effect of j th level of factor B. (αβ)ij denotes the interaction effect of ith level of factor A and
jth level of factor B yijk represents the kth observation corresponding to ith level of factor
A and jth levle of factor B. µ is the general mean effect € ijk ‘s are independent and normally
distributed error terms.
. The effect underlying assumptions and hypothesis to be tested do differ for the three
models.
Fixed model
It can be shown that ∑αi = ∑βj = 0 ∑(αβ)ij when summed over I or j or both equals zero.
2
€ijk -> IND(0, σ )
H01 : αi =0 fo r i=1, 2, …….a H11 : αi ¿ 0 at least for some I
H02 : βj =0 for j-1,2……b. H12: βj ¿ 0 at least for some j
H03 : (αβ)ij = 0 for I= 1,2, ……. a and j=1,2……..b H13 : not all (αβ)ij are zero.
Random effect model
2
Here αi IND(0, σ α )
Βj I N(0,
σ 2β ) (αβ)ij IN(0,
σ 2αβ ) 2
€ijk = IN(0, σ )
2 2
H01 : σ α =0 H11: σ α ¿ 0
2 2
H02 : σ β =0 H12 : σ β ¿ 0
2 2
H03 : σ αβ =0 H13: σ αβ ≠¿ ¿0
Mixed effect model
A fixed B random
∑αi =0
Βj I N(0,
σ 2β ) (αβ)ij IN(0,
σ 2αβ ) 2
€ijk = IN(0, σ )
H01 : αi =0 for i=1, 2, …….a H11 : αi ¿ 0 at least for some i
2 2
H02 : σ β =0 H12 : σ β ¿ 0
2 2
H03 : σ αβ =0 H13: σ αβ ≠¿ ¿0
Example
A study is performed on the productivity of employees in a large manufacturing company.
Company wants to get an idea of daily productivity and how it depends on which machine
an employee uses. m employees and r machines having each employee work on each
machine for a total of n days. If these are only the employees and these are only the
employees , ie, m machine is not a sample from a large population of employees and r
machine does not represent a sample from large population of machines , then this we call a
fixed effect model and will be analysed under fixed effect model. On the other hand if m
employees represent a random sample from a large sample of population of employees and
r machine is a random sample from a large population of machines then this will be a
random effect model and will be analysed under random effect model. Any conclusion
drawn on the basis of r machines will be valid to entire population of machines and m
employees will be valid to entire population of employees. If m employees is not a random
sample from large population of employees but r machines represent a random sample
from large population of machines, then the corresponding model defines a mixed effect
model and will be analysed under mixed effect model.