You are on page 1of 3

REGRESSION:

Linear regression Logistic regression


Outcome: continuous Outcome: binary (categorical)
For quantifying the relationship between predictor and outcome => clinical
equation
Simple Simple
=> gives the crude OR
Steps: Steps:
1- Model significance (hypo) 1- Model’s significance (hypo)
2- Independent t test - Fitted model – interpretation 2- Independent t test – interpretation
3- Confidence interval 3- Confidence interval
4- R squared

Hypo: Hypo:
H0: β=0 H0: OR = Exp β=1
H1: β≠0 H1: OR = Exp β≠1
Model’s significance: Model’s significance:
Look at t-test’s p-value: if < 0.05 Look at t-test, the OR= Exp β and its p-value: if <0.05
=> we reject H0 and deduce that at alpha =0.05 level of significance the => we reject the null hypothesis of no association and deduce that at
model is significant. There is an association between the variable and the alpha=0.05 lvl of significance, the variable is a predictor of the outcome.
outcome. The variable is a predictor of the outcome.

Fitted model: Interpret the crude OR


Y=α+βX
Y, alpha and beta hat
Interpretations: Interpretations:
Predictor: continuous Predictor: binary Predictor: continuous Predictor: binary
If β>0: If β>0: If β>0: If β>0:
When the predictor increases by 1 When the predictor moves from 0 When the predictor increases by 1 The odds of getting the outcome
unit, the outcome increases by β to 1, the outcome increases by β unit, the odds of the outcome for the predictor coded 1 is OR=Exp
units on average. units on average. increase multiplicatively by OR=Exp β times higher than that for the
β times. predictor coded 0.

If β<0: If β<0:
If β<0: If β<0: When the predictor increases by 1 The odds of getting the outcome
When the predictor increases by 1 When the predictor moves from 0 unit, the odds of the outcome for the predictor coded 0 is OR=Exp
unit, the outcome decreases by β to 1, the outcome decreases by β decrease multiplicatively by OR= β times higher than that for the
units on average. units on average. Exp β times. predictor coded 1.
or: when the predictor decreases or: when the predictor moves from
by 1 unit, the outcome increases by 1 to 0, the outcome increases by β
β units on average. units on average.
CI CI CI CI
We are 95% confident that the true We are 95% confident that the true We are 95% confident that the true We are 95% confident that the a
effect of a one unit increase in the effect of a 0 to 1 move in the OR for the outcome coded 1 for the one unit increase in the predictor
predictor on the outcome denoted predictor on the outcome denoted variables coded 1 versus 0 ranges leads to a true OR for the outcome
as β ranges between – and –. as β ranges between – and –. between – and –. coded 1 that ranges between – and
–.

This CI does not cross 0, we reject the null hypothesis and deduce that at This CI does not cross 1, we reject the null hypothesis and deduce that at
alpha = 0.05 the variable is a significant predictor of the outcome. alpha = 0.05 the variable is a significant predictor of the outcome.
Both limits are: Both limits are:
>0 – the relationship between the predictor and the outcome is >1 – the variable increases the risk of the outcome, therefore the variable
incremental, when the predictor increases, the outcome increases. is a risk factor.

<0 – as the predictor increases, the outcome decreases. <0 – the variable decreases the risk of the outcome, therefore the variable
is a protective factor.
OR:
This CI crosses 0, we fail to reject the null hypothesis and deduce that at OR:
alpha = 0.05 the variable is not a predictor of the outcome. This CI crosses 1, we fail to reject the null hypothesis and deduce that at
alpha = 0.05 the variable is not a predictor of the outcome.
OR>1 -> harmful factor
OR<1 -> protective factor

To express a protective factor as a harmful factor = 1/OR(protector)


R squared:
- coefficient of determination

R squared % of the variability in the outcome is explained by the


predictor/s.

NB: If % > 90-95 => v good predictive model.

Interpret the slope β:


Look at the fitted model:
β>0 – increase
β<0 – decrease

As predictor increases by 1 unit, outcome increases by β units on average.

Multiple Multiple
=> gives the adjusted OR

Steps: Steps:
1- Model significance (hypo) – F test 1- Model’s significance (hypo)
2- Separate t tests for each predictor – fitted model - 2- Separate independent t test – interpretation
interpretation 3- Confidence interval
3- Confidence interval Hypo:
4- R squared H0: All ORs = 1
Hypo: H1: at least one OR is ≠1
H0: β(a)= β(b)= β(c)=0
H1: at least 2 of the β is ≠0
Model’s significance: Model’s significance:
Look at the overall F test statistic, its p-value: if <0.05: Look at p-values of the ORs, if <0.05
=> we reject the null hypothesis of no association and deduce that at => we reject the null hypothesis and we can deduce that at alpha=0.05, the
alpha=0.05, at least one of the variables is significantly associated with the model is significant and at at least one of the ORs is different than 1.
outcome.

Separate t-test for each predictor: regression table Separate t-test for each predictor:
Hypo:
H0: β(a)=0 in a model that contains b and c
H1: β(a)≠0 in a model that contains b and c

Look at t statistic value, and p-value, if <0.05


=> we reject H0 and deduce that at alpha=0.05, in a model that contains b
and c, A is a significant predictor of the outcome.

Fitted Model:
Y = α + βa X + βb X + βc X

Interpretations:

Predictor: continuous Predictor: binary


If β>0: If β>0:
Adjusting for b and c, when the Adjusting for b and c, when the
predictor a increases by 1 unit, the predictor a moves from 0 to 1, the
outcome increases by β units on outcome increases by β units on
average. average.

If β<0: decreases If β<0: decreases

CI CI
Adjusting for the effect of b and c, Adjusting for the effect of b and c,
we are 95% confident that the true we are 95% confident that the true
effect of a one unit increase in the effect #of a 0 to 1 move in the
predictor a on the outcome predictor on the outcome denoted
denoted as β ranges between – and as β ranges between – and –.
–.
This CI does not cross 0, we reject the null hypothesis and deduce that at
alpha = 0.05 the variable a is a significant predictor of the outcome in a
model that contains b and c.
Both limits are:
>0 – the relationship between the predictor and the outcome is
incremental, when the predictor a increases, the outcome increases in a
model that contains b and c.

<0 – as the predictor a increases, the outcome decreases in a model that


contains b and c.

OR:
This CI crosses 0, we fail to reject the null hypothesis and deduce that at
alpha = 0.05 the variable is not a predictor of the outcome in a model that
contains b and c.
R squared:
- coefficient of determination

R squared % of the variability in the outcome is explained by the predictors

NB: If % > 90-95 => v good predictive model.


Interpret the slope β:
Look at the fitted model:
β>0 – increase
β<0 – decrease

Adjusting for the effect of b and c, as predictor a increases by 1 unit,


outcome increases by β units on average

You might also like