Help with scatterplots for a panel

[Mario Ferri]

Hello


I am trying to get some scatterplots for a panel of variables that include also some indicators, derived from categorical variables.When I am trying to get the graph the indicators variables are showing as lines verdical or horizontal. In theory and in previous works they are supposed to show a correlation (strong or other). I know the most common answer will be to look in my data but still..

The command I am running is

graph matrix  varlist

Any help on this?

[Mario Ferri]

Hello


I am trying to get some scatterplots for a panel of variables that include also some indicators, derived from categorical variables.When I am trying to get the graph the indicators variables are showing as lines verdical or horizontal. In theory and in previous works they are supposed to show a correlation (strong or other). I know the most common answer will be to look in my data but still..

The command I am running is

graph matrix varlist

Any help on this?

I have winsorized the data using winsor2 today in order to try to get better results but still . Any ideas please?

[Nick Cox]

If any variable is 0 or 1 only then it can only plot on two vertical lines on an x axis and on two horizontal lines on a y axis. That doesn't rule out either positive or negative correlations: all depends on the distribution of the variables. In fact for two such variables you can get correlations of 1 or -1 if only opposite corners of a scatterplot have data points and high correlations in many other cases.

[Mario Ferri]

If any variable is 0 or 1 only then it can only plot on two vertical lines on an x axis and on two horizontal lines on a y axis. That doesn't rule out either positive or negative correlations: all depends on the distribution of the variables. In fact for two such variables you can get correlations of 1 or -1 if only opposite corners of a scatterplot have data points and high correlations in many other cases.

While I cannot share data I upload an image of the scatterplot . I have no 0 or 1 in the dataset. The variables of interest are pm, gv and sm and their relations with the other variables.I winsorized the data using winsor2 in order to try to get better results but still .Comments and suggestions are welcome!


Thanks

[Nick Cox]

You mean something different by indicator. OK, but now I fail to understand what you seek here or even what better results would mean. Sorry.

[Mario Ferri]

You mean something different by indicator. OK, but now I fail to understand what you seek here or even what better results would mean. Sorry.

I mean another type of scatter plots or a way to manipulate data in order to see their graphics relation with the other variables in ie a scatter plot. Sorry for not being clear earlier

[Nick Cox]

Difficult to say. You may have lots of exact zeros. Logarithmic scale might help.

[Mario Ferri]

Difficult to say. You may have lots of exact zeros. Logarithmic scale might help.

Many thanks for this.Taking logs works well!