Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

 From Shikha Sinha To statalist@hsphsun2.harvard.edu Subject Re: st: cascading dummies Date Mon, 1 Oct 2012 16:18:45 -0700

```Richard,

You and also STB  mentioned that coeff in regression using ordinal
dummies can be constructed by subtracting the coeff from the
regression using the standard dummies. Is this true for non-linear
regression as well (Logit, probit)? I could get it that way in an
ordinary least square but not sure about the non-linear regression.

Also, how do we interpret the coeff from cascading dummies?

lit | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
nfac1_12 |   1.674494   .0183205    47.12   0.000     1.638969     1.71079
nfac1_13 |   1.546588   .0202219    33.35   0.000     1.507457    1.586734
nfac1_14 |   1.397527   .0223861    20.90   0.000     1.354332    1.442099
nfac1_15 |   1.579932   .0347163    20.82   0.000     1.513334    1.649462

Does the odds ratios of 1.397527 mean that an increase in the dummy
from 3 to 4 results in 39% increase in literacy?

Thanks,
Shikha

On Mon, Oct 1, 2012 at 12:15 PM, Nick Cox <njcoxstata@gmail.com> wrote:
> As Richard said, Stata can tell you:
>
>
> STB-6   srd11 . . . . . . . . . Generating ordered "cascading" dummy variables
>         (help cascade if installed) . . . . . . . . . . . . . . . R. Goldstein
>         3/92    pp.19--22; STB Reprints Vol 1, pp.190--194
>         add a set of new dummy variables that are cascading; that is,
>         0/1 variables are created where a 1 is given if the case has
>         the value <= value of lowest category
>
> Accessible at http://www.stata.com/products/stb/journals/stb6.pdf
>
> Nick
>
> On Mon, Oct 1, 2012 at 7:59 PM, Shikha Sinha <shikha.sinha414@gmail.com> wrote:
>> Thanks Richard!
>>
>> very helpful. What is the full reference of STB article, I am unable to find it.
>
> On Mon, Oct 1, 2012 at 11:39 AM, Richard Goldstein
>> <richgold@ix.netcom.com> wrote:
>
>>> neither is wrong or right -- they answer slightly different questions; A
>>> asks for each dummy whether, and by how, much it is different from the
>>> reference group (you do realize that you should only include 4 of the
>>> dummies, right?); B asks whether each differs from the preceding level
>>>
>>> you can -findit cascade- to find a program I wrote to implement
>>> cascading dummies; the help file, and even more the STB article,
>>> discusses the differences; note that you can obtain the answer to either
>>> question by following up the original method of forming the variables
>>> with the appropriate -test- command(s)
>
> On 10/1/12 2:31 PM, Shikha Sinha wrote:
>
>>>> Recently, I came across a new way of creating dummies and I wonder
>>>>
>>>> The independent variable X is coded as 1- very poor, and 5 as very
>>>> rich. I want to estimate the effect by wealth quintile. I created the
>>>> dummy the following ways, but I was told that this is wrong (A is
>>>> wrong). The correct way to construct dummy is B and is called
>>>> cascading dummies. I have never come across this before and would
>>>> appreciate if you could shed light on the difference between the two
>>>> and which is the correct way of creating dummies.
>>>>
>>>> A:
>>>> id    Y       X1 (scale of 1-5),      dum1    dum2    dum3    dum4    dum5
>>>> 1     100     5       0       0       0       0       1
>>>> 2     200     4       0       0       0       1       0
>>>> 3     300     3       0       0       1       0       0
>>>> 4     239     2       0       1       0       0       0
>>>> 5     345     1       1       0       0       0       0
>>>>
>>>>
>>>> B:
>>>> id    Y       X1 (scale of 1-5),      dum1    dum2    dum3    dum4    dum5
>>>> 1     100     5       1       1       1       1       1
>>>> 2     200     4       1       1       1       1       0
>>>> 3     300     3       1       1       1       0       0
>>>> 4     239     2       1       1       0       0       0
>>>> 5     345     1       1       0       0       0       0
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/
```