MATLAB: N-way ANOVA giving NaN and Inf

n-way anova anova

Hello,

I tried to performe a n-way anova with the dataset in attachement. Also in attachement you can find Data_Read_2003_2010.m to import the dataset into matlab.

If I use only two terms and one factor I can get p-values for each of the terms tested.

Once I try to use three or more terms, I get the values from the print screen bellow.

The command I am executing is the following: [p,tbl2,stats]=anovan(BPI_r,{GerBill' Yield_Spread_DE' PT_DE_Spread'})

Anyone know why I am getting Inf and NaN for F and p-values when I am using three or more terms in n-way anova? I also trying with different terms in the dataset and the entire dataset.

Best regards,

Bruno

Best Answer

According to the documentation, "By default, anovan treats all grouping variables as fixed effects." Your grouping variables do not seem to fit that description. They are continuous variables rather than factors.

% If your variables were factors, these plots would produce staircases
clf()
subplot(3,1,1)
plot(sort(GerBill))
subplot(3,1,2)
plot(sort(Yield_Spread_DE))
subplot(3,1,3)
plot(sort(PT_DE_Spread))

You'll need to indicate that they are continuous predictors by using the continuous property.

 [p,tbl2,stats]=anovan(BPI_r,{GerBill' Yield_Spread_DE' PT_DE_Spread'},'continuous',1:3)

Result

You were getting NaNs because you need to have at least two replications for the same set of factor levels and since your "factors" were in reality continuous variables, there were not any repeated levels.

Related Solutions

MATLAB: Multivariate analysis of variance in Matlab

You need to arrange your data in a table with one row per individual group member. For example, your first group seems to have about 55 members that will go into 55 rows, your second group has about 133 members (133 rows), and so on. So, you need a table with about 700-800 rows, as the sum of the different group sizes.

In the required table, you need one column for each dependent variable, as you have now, plus you need one more column with values indicating group membership, e.g., values like "GroupA", GroupB", etc.

It isn't actually possible to produce the required table from the *.mat file that you posted, because it appears that you have some missing data. For example, in the first row (group) some variables have 55 scores but others have only 53 or 54. That makes it look like there were 55 members in that group but scores were lost on a few variables for a few of the group members. The manova depends on the correlations among variables, so the data set has to make it clear how the scores of different variables match up. This depends on which of the (say, 55) first group members were missing data on variables 2 and 7, for example, and there is no way to tell that from the posted mat file.

MATLAB: Correlation between two row matrices

Like that, each value of "a" is correlated to each value of "b", but applying the formula of the correlation, the correlation of two single numbers is NaN. To compute the correlation correctly, traspose the input vectors

result  = corr(a', b');

Best Answer

Related Solutions

MATLAB: Multivariate analysis of variance in Matlab

MATLAB: Correlation between two row matrices

Related Question