Beyond ANOVA (Part-5)

[Neeraj Kaushik]

Dear Friends

Lets work with the rest of the options of ANCOVA:
1. Model
2. Contrast
3. Plot
4. Post hoc (This become unavialble in as soon as we enter any var in Co-variate)
5. Save
6. Options
7. Bootstrap

1. Model:

The default modle if Full factorial which involves measuring the effect of all indep var as well as their interactions.
In Ancova the effect of co-variate is neutralized, hence it wont be the part of any interaction with indep var.

We prefer working with deafult model Type-III, ehnce no change here. Type-IV is used when there are many missing values.

 

2. Contrast

It is same as Post-hoc. Now as the name 'Post hoc' indicates it is to be performed after the analysis. Whether to use Contrast (the same as Post hoc) is decided before analysis. Another thing is Post hoc performs paired comparison in all categories of indep var while Contrast perform compaison NOT in all pairs rather between the 'first & the rest all' or 'last & rest'

For begining we shd avoid using it. I'll write about it in detail later.

3. Plot

Plot is a gud way of judging the things. But plots are made only with the indep var & never for the co-variate.

The dep var is selecetd by default.

To begin with we put indep var in Horizontal axis and must click on ADD & then Continue.

 

4. Post Hoc

In case we're not using co-variate, then we shd use Post hoc by Tukey method. 

 

5. Save

Two things we may be interested in saving are: Predicted values & 2nd is residual.

To begin with, lets not use it.

 

6. Options:

This one is the most used menu:

On Top is given Marginal Estimated means. Input the indep var (for which we want to see the mean scores) on left hand side pane.

Now interesting thing here is that these estimated means are different from actual means (as given by Descriptive under Display).

So Marginal Estimated means are mean when effect of co-variate is taken care of.

 

Under Display:

Use  Descriptive score (It'll give the mean score of dep var on the basi sof different categories)

Use Estimates of the effect size (This will give partial eta squared which is equivalent to the r-sq of Regression)

Other options can be avoided in beginning.

 

 
Regards
Neeraj

[Radha garg]

Respected Sir,

Thanks for teaching us the ANCOVA. As we know that Ancova is a extension of anova so it is also a parametric test which is use when the data is normally distribution. But most of times our data is not normally distributed or skewed. So, is there any alternative for ANCOVA in non parametric test.

Though i tried to search it on net but couldn't find the answer.

So, pls tell something about its alternative test, if it's available.

Thanks and Regards

Radha

[Neeraj Kaushik]

Dear Radha

The technique which u r looking is Mulyi Nominal regression.

I'll write on it shortly. In the meanwhile u can read its tutorial in SPSS from Help-->Show me option.

Best wishes

Neeraj