Graphing a cubic model in stata

[Daniel Hamill]

Hi I am trying to graph several cubic functions on one graph in stata (or even one at a time) based on my estimated regression. My regression is in the form y = x + x^2 + x^3 + z + z^2 + z^3 + v + v^2 + v^3 + w + w^2 + w^3 + binary variables.

I have looked at examples such as: http://www.stata.com/statalist/archi.../msg00766.html, which is a question from the old help form. The example they use is:
sysuse auto, clear reg mpg c.weight##c.weight##c.weight predict mhat sc mpg weight||line mhat weight, sort This works when estimating one cubic but I want to do 4 at a time and graph them individually or collectively with error bars. I don't need to plot the residuals as I have 20,000 + data points

my data is in panel form.
To summarize how do you graph cubics from a regression estimation against the y value.

Any help is much appreciated. If you need any more specific information let me know as I tend to miss important information occasionally.

Cheers,
Daniel

[Oded Mcdossi]

You can use the margins command to plot each curve and its CI's in a separate graph. If all of your x, z, v and w variables are of the same scale you can, I guess, plot them in one graph, though, I think it would be too messy. If your variables are of different scales than you can rescale it to SD units or refer to SD units at the margins command and than plot it in the same graph.

[River Huang]

Hi I am trying to graph several cubic functions on one graph in stata (or even one at a time) based on my estimated regression. My regression is in the form y = x + x^2 + x^3 + z + z^2 + z^3 + v + v^2 + v^3 + w + w^2 + w^3 + binary variables.

I have looked at examples such as: http://www.stata.com/statalist/archi.../msg00766.html, which is a question from the old help form. The example they use is:
sysuse auto, clear reg mpg c.weight##c.weight##c.weight predict mhat sc mpg weight||line mhat weight, sort This works when estimating one cubic but I want to do 4 at a time and graph them individually or collectively with error bars. I don't need to plot the residuals as I have 20,000 + data points

my data is in panel form.
To summarize how do you graph cubics from a regression estimation against the y value.

Any help is much appreciated. If you need any more specific information let me know as I tend to miss important information occasionally.

Cheers,
Daniel

The command

marginscontplot

might help. (Please search "marginscontplot".)

[Daniel Hamill]

before I can use marginsplot command I need to use margins first. Could you please provide an example of code using this function?:

xtreg y c.x##c.x##c.x c.v##c.v##c.v c.z##c.z##c.z c.w##c.w##c.w

where they are all continuous variables.

Cheers,
Daniel

[Clyde Schechter]

You have to select appropriate values of x, v, z, and w, values which will adequately cover the range of values you are interested in, and enough in between so that the resulting line-graph will look adequately curvilinear. Suppose you are interested in values of x between 0 and 10. Then you can do:

margins, at(x = (0(0.25)10))
marginsplot

and similarly for each of the others.

If you really want to plot them all on one graph, then,:

margins, at(x = (0(0.25)10)) at(v = (v values)) at(z = (z values )) at (w = (w values)) saving(for_graphs, replace)
use for_graphs, clear

and then write your own -graph twoway- command using the results in for_graphs.dta

[River Huang]

before I can use marginsplot command I need to use margins first. Could you please provide an example of code using this function?:

xtreg y c.x##c.x##c.x c.v##c.v##c.v c.z##c.z##c.z c.w##c.w##c.w

where they are all continuous variables.

Cheers,
Daniel

Please try:

use "grunfeld.dta", clear

egen id = group(firm)
xtset id year

xtreg invest c.value##c.value##c.value c.capital##c.capital##c.capital
marginscontplot value
marginscontplot capital

[Daniel Hamill]

Thanks Clyde Schechter My graphs look fantastic thanks to the commands below.
margins, at(x = (0(0.25)10)) marginsplot I ran the over code you noted:
margins, at(x = (0(0.25)10)) at(v = (v values)) at(z = (z values )) at (w = (w values)) saving(for_graphs, replace) use for_graphs, clear But got the error "file for_graphs.dta could not be opened". Any idea what I would be doing wrong? is this creating a temporary file? which is then opened? Again thanks for the awesome help so far. Cheers, Daniel

[Clyde Schechter]

I don't know what's wrong there. I've used this construction myself in the past without difficulty. Which command is throwing the error message, the -margins- command or the -use- command?

One possibility: the -saving()- option of -margins- is undocumented. If you are not running the latest version of Stata (14) it may not even be implemented, or not correctly implemented in your version of Stata.

The code as written does not create a tempfile--it should create a permanent file named for_graphs.dta. If you look for it in your working directory, do you find it there? Do you have full access privileges on the directory you're working in? Is the code you showed in #7 exactly what you ran? Any chance that for_graphs is misspelled in either place? (And count change of upper and lower case as misspelling if you are on a Mac.)

In the future, when showing code you ran, please put it between code delimiters as I have done. Code that is run together on one line is difficult to read. It also forces me to assume that your line breaks are where they should be, and not somewhere else causing errors.

[Daniel Hamill]

Hi when I use the margins command below the margins are produced but I get the message at the end:

"(note: file for_graphs.dta not found), file for_graphs.dta could not be opened,r(603)"

"

margins, at(x = (5(5)180)) at(y = (5(5)180)) at(z = (5(5)180)) at (w = (5(5)180)) saving(for_graphs, replace)

I am using stata 13. I couldn't see it in my working directory.

[Clyde Schechter]

I don't know what to tell you. Perhaps version 13 just doesn't support the -saving()- option with -margins-. Are you sure you have write privileges in your working directory? Do other -save- commands or -saving()- options produce saved files when you run them in this directory?

[Bruce Weaver]

Clyde wrote in #10:

I don't know what to tell you. Perhaps version 13 just doesn't support the -saving()- option with -margins-.

I have version 13.1, so typed help margins, searched for sav in the Help window, and came up empty.

Cheers,
Bruce

[Andrew Musau]

An alternative to saving the predicted margins is through the user written command -marginscontplot- by Patrick Royston. I recently replied to a post here, so see #3 in the following for details

http://www.statalist.org/forums/foru...ine-mcp-graphs

Of course, upgrading to Stata 14 is always recommended

[Oded Mcdossi]

Saving under margins command in Stata 13 is an undocumented option but it is working. For more details type help margins_saving.

[Daniel Hamill]

Hi Clyde, I tried running this same function on another computer with stata 14 and got the same message that "file for_graphs.dta not found". I should have access to my computers directory. Can I choose to save this not in my directory if I don't have access? I would usually save else where and am not familiar with the saving option on commands.

[Clyde Schechter]

You can specify a full pathname in the -saving()- option.

But let's back up a second. When are you getting this error message? Are you getting it after -margins-, or when you try to -use for_graphs-? If the former, it's not a problem. Whenever you save a file with the -replace- option, Stata gives you a note indicating that the file was not found if that file did not already exist there. It's basically just checking in with you in case you were thinking the file already existed. But unless the file should have already existed, it's not a problem.

If, however, you're getting -file not found- when you try to -use- that data set, then there is a problem in that the file is missing when it should be there.