Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: 1-4 scale
David Hoaglin <firstname.lastname@example.org>
Re: st: 1-4 scale
Sun, 5 Aug 2012 12:59:34 -0400
Good points. Fortunately, the variables that Ebru is asking about are
potential predictor variables.
On Sun, Aug 5, 2012 at 1:52 PM, Richard Williams
> At 10:57 AM 8/5/2012, David Hoaglin wrote:
>> Dear Ebru,
>> People often analyze data from Likert scales as equally spaced, so you
>> can use each of the eight items in your model as a numerical variable,
>> with values 1 to 4. You simply need to be aware that you are treating
>> the four categories as equally spaced.
> It is a debatable practice though. Consider the following (warm has 4
> use "http://www.indiana.edu/~jslsoc/stata/spex_data/ordwarm2.dta", clear
> reg warm yr89 male white age ed prst
> According to the reference manual discussion of rvfplot, "In a well-fitted
> model, there should be no pattern to the residuals plotted against the
> values...Any pattern whatsoever indicates a violation of the least-squares
> Clearly, there is a pattern in the above rvfplot, i.e. you get 4 parallel
> straight lines. Further, it isn't unique to this example; any 4 category
> dependent variable will show the same thing.
> In fairness, if your dependent variable had 17 possible values, you would
> have 17 straight lines -- but your eye probably wouldn't detect that because
> everything would seem so cluttered. There is probably some point where there
> are enough possible values that violations of OLS assumptions aren't
> important, but I would be hesitant to say that point is met with a DV that
> only has 4 categories.
>> Earlier you asked about centering those variables. Centering will do
>> no harm. As far as the model is concerned, it affects only the
>> definition of the intercept. If you do decide to "center" the
>> variables, you may want to use one of the four values. If the data on
>> an item are not concentrated at one end, you could use 2 or 3 or
>> perhaps 2.5 as the centering constant. (In a 5-point Likert scale
>> with a neutral category at 3, using 3 would often be a reasonable
>> When you have the results from the model with the eight separate
>> items, you may want to see whether the coefficients for the four items
>> within a heading are similar. If they are, and it makes sense, you
>> could consider replacing those four items with their sum (or average)
>> --- a composite score.
>> David Hoaglin
* For searches and help try: