Re: st: plotting a regression function with time-dummies indicating structural breaks

[Nick Cox <njcoxstata@gmail.com>]

-twoway function- has been around since Stata 8 (2003). Its use of x
regardless of whatever x might be in the data is documented. Its
capacity for small surprises when you want to use some or all of your
data as well is less surprising on brief examination. I agree that
there is a case for considering a tweak of the documentation here, but
I see no evidence that -twoway function- does not work as intended.

Nick

On Tue, May 10, 2011 at 2:25 PM, Allan Reese (Cefas)
<allan.reese@cefas.co.uk> wrote:
> Back in March I had a similar problem to Oliver Jones and discussed
> off-list with Nick Cox
> (see below, reading from end). We let it drop, but it clearly confuses
> more users than just
> me that "graph twoway function" operates without the data matrix.  Using
> names of variables
> in the function to be plotted produces unexpected results.
>
> Maybe the developers could look at the parsing and issue warning
> messages, and at the
> documentation to clarify that italic "x" is a literal and not (like
> "exp" and "range")
> a token to be substituted. The better approach, when the X axis is a
> data variable, is
> Maarten's suggestion of generating points on the line and -connect-ing
> them.
>
> Allan
>
> R Allan Reese
> Senior statistician, Cefas
> The Nothe, Weymouth DT4 8UB
>
>
>
> --- Oliver Jones asked (statalist 3 May)
>>> I wonder if there is an easy way to plot a regression of the form:
>>>
>>> y_t  b_0 + b_1*TIME_t + Dummy_1*(b_2 + b_3*TIME_t) + b_4*Dummy_2
> [where b_0 etc are scalars and Dummy_1 etc are variables]
>>>
>>> So far I tried:
>>> gen x  TIME_t
>>> twoway ///
>>>  (tsline y) ///
>>>  (function y  b_0 + b_1*x + Dummy_1*(b_2 + b_3*x) + b_4*Dummy_2, ///
>>>  range(`first_year' `last_year'))
>
> & Maarten buis replied:
>> I tend to plot such graphs by first using -predict- or -adjust- to
>> get (adjusted) predictions and than use -twoway line-. It is just
>> too easy to make a typo in this type of -twoway function- calls.
> -------------------------------------------------------------------
>
>
>
>
> -----
> From: Allan Reese (Cefas) [mailto:allan.reese@cefas.co.uk]
> Sent: 18 March 2011 13:50
> To: Nick Cox
> Subject: RE: Bug in "twoway function" (v 11.1 used)
>
> It's a nice exercise for the little grey cells - years of training that
> f(x) is just a function
> led to assumption I should substitute the "x" variable in the function
> call. Expression parsing
> should allow the program to determine that f(x) with no x term should be
> constant. As it's not
> necessarily evaluating f() where the data has values and -price- is not
> a function, I would agree
> with your last point.
>
> Allan
>
> ----
> From: Nick Cox [mailto:n.j.cox@durham.ac.uk]
> Sent: 18 March 2011 13:36
> To: Allan Reese (Cefas); 'Stata Technical Support'
> Subject: RE: Bug in "twoway function" (v 11.1 used)
>
> A better answer, perhaps, is that Stata takes your RHS in
>
> [LHS ] RHS
>
> and evaluates it, after substituting for any x with a grid of points on
> [0,1].
>
> If you rescale the range, it does so too.
>
> The surprise is that it doesn't actually throw you out if there is no
> reference to x.
>
> Nick
> n.j.cox@durham.ac.uk
>
> ----
> From: Nick Cox
> Sent: 18 March 2011 13:16
> To: 'Allan Reese (Cefas)'; Stata Technical Support
> Subject: RE: Bug in "twoway function" (v 11.1 used)
>
> I think the question is whether this is Stata's misfeature or yours.
>
> Perhaps -twoway function- should reject your syntax whenever it doesn't
> use -x-, which is
> explicitly expected, but it does the best it can to make sense of what
> you said. The effect
> appears to be equivalent to
>
> gen obs  _n
> scatter price obs
>
> but scaled to the range of price.
>
> Evidently your syntax is not illegal, otherwise it would have been
> rejected, but it does not
> produce anything useful.
>
> I think this is on the level of if you ask a strange question, you may
> get a puzzled reply!
>
> Nick
> n.j.cox@durham.ac.uk
>
> ----
> From: Allan Reese (Cefas) [mailto:allan.reese@cefas.co.uk]
> Sent: 18 March 2011 11:11
> To: Stata Technical Support
> Cc: Nick Cox
> Subject: Bug in "twoway function" (v 11.1 used)
>
> Documentation for -twoway function- states:
> You type -twoway function ysqrt(x)-
> It makes no difference whether y and x are variables in your data.
>
> Unfortunately it does. Using the auto data, compare
>
> . scatter weight price || function yx, range(price) n(20)
>
> . scatter weight price || function yprice, range(price) n(20)
>
> . scatter weight price || function yprice, range(price)
>
> Copy price as "x" and plot yx and it works, so what was meant appears
> to be that you -must- define the f(x) in terms of the dummy "x" and a
> variable called "x" in the dataset will not be used. But it is not
> obvious to me how price is used in the second and third examples. As
> there is no x, the function ought to be constant.
>
> It's also confusing to write that the "... in range play[s] no part
> unless option range(varname) is specified." The -in range- selects
> cases.
>
> Of course, only an idiot would not understand f(x)...
> Allan
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/