Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: Graphs With Log Scale: A Bug?


From   "Nick Cox" <[email protected]>
To   <[email protected]>
Subject   RE: st: Graphs With Log Scale: A Bug?
Date   Thu, 29 May 2008 14:38:36 +0100

Glad the advice was helpful. 

Nick 

Vladimir V. Dashkeyev

Nick,

I understand that "make it work for me" approach is not acceptable.
But before asking I made several attempts and failed. I do appreciate
your help, I managed to get what I wanted thanks to your advice.

Thank you again,
Vladimir

On Wed, May 28, 2008 at 5:48 PM, Nick Cox <[email protected]> wrote:
> Please accept more responsibility for solving problems.
>
> It is just a matter of (a) reading the help and (b) applying textbook formulae.
>
> Here is a sketch:
>
> local level = <your_choice, e.g. 95>
> regress <whatever>
> tempvar pred se ul ll
> predict `pred'
> predict `se', stdp
> local level = (100 - `level') / 200
> gen `ul' = `pred' + invttail(e(df_r), `level') * `se'
> gen `ll' = `pred' - invttail(e(df_r), `level') * `se'
>
> Nick
> [email protected]
>
> Vladimir V. Dashkeyev
>
> Nick,
>
> Thanks for the answer. I did not use -predict- since this approach
> does not provide a quick way for drawing confidence intervals. If I'm
> wrong and there is a way to draw CI, please, let me know about it.
>
> Thank you,
> Vladimir
>
> On Wed, May 28, 2008 at 4:21 PM, Nick Cox <[email protected]> wrote:
>> My advice is to use -predict- after each model fitted to save the results in separate variables. Then draw one graph to get you want. I wouldn't approach this via -lfit- or
>> -lfitci-. That will also oblige you to make explicit what you are doing.
>>
>> Nick
>>
>> Vladimir V. Dashkeyev
>>
>> Thanks for the reply. I should have emphasized in the first message,
>> that I run -lfitci- of X on ln(Y) in both scenarios. The difference is
>> in the scatter plot. In the first scenario I use ln(Y), and in the
>> second -- Y with log scale option. I expected to get the same linear
>> prediction line and the same scatter plot.
>>
>> But after I posted that question, I compared the graphs once again and
>> realized that the real problem is with the Y axis scale. If I draw a
>> scatter and prediction line on the same Y axis, everything is fine.
>> Yet if I draw the same scatter with 2 Y axes I get different range of
>> values on Y1 and Y2 axes. I need two Y axes for overlaid drawing of
>> the scatter with -yscale (log)- option and linear prediction of
>> X-ln(Y). Setting range on both axes to the same values did not help.
>> They are very close but still shifted a bit. So the arrangement of
>> observations and prediction line is not correct. So it's not a bug,
>> but still a problem I have to solve.
>>
>> Is there any way to "tie" axis Y1 with axis Y2?
>>
>> Maarten buis
>>
>>> --- "Vladimir V. Dashkeyev" <[email protected]> wrote:
>>>> I drew a two-way plot with a linear prediction line -lfitci- of X on
>>>> natural logarithm of Y. Next, I drew the plot of X on Y with log
>>>> scale option -yscale(log)-.
>>>>
>>>> To my surprise regression line changed its slope. The slope is
>>>> greater with the -yscale(log)- option. I used the same X axis and
>>>> the second Y-axis for the linear prediction graph .
>>>> Is this a bug or am I doing something wrong?
>>>
>>> This is not a bug: in the first scenario you are thinking that there is
>>> a linear relationship between ln(Y) and X and you are showing the
>>> predictions, while in the second scenario you are thingking that there
>>> is a linear relationship between Y and X and then transforme the
>>> predictions to a log scale. So the results are different because the
>>> models are different.

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index