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

Re: st: RE: Your opinion on income groups and inflation

Subject   Re: st: RE: Your opinion on income groups and inflation
Date   Sat, 7 Jun 2008 14:42:13 -0400

On income groups (intervals), I would not use dummies because you have
information about income _levels_  which would be otherwise lost. An income
interval of 300 to 400, is not the same thing as an income interval of 1200 to
3600. Since you do not have information about distribution of income within
intervals, nor apparently about the means, I would use the average of the
interval as my income variable. (Assumes, of course, a uniform distribution
within the interval. If you want to be "fancier", you may want to use something
like the 65th percentile of each interval on the assumption that the
distribution is lognormal within each interval, and thus the mean is to the
right of the median; empirical income distributions have a mean which normally
lies btw. the low 60s and low 70s percentiles).

On trust, you may want to use dummies--on the assumption that moving from trust
level 1 to 2 is not the sane thing than moving from trust level 8 to 9 etc. But,
on the other hand, keeping an ordered categorical variable makes it easier to
interpret the results, and is easier on the reader who does not need to be
presented with ten variables.

Best, Branko

Development Research, World Bank
Email: or branko_mi@yahoo.
tel: 202-473-6968
World Bank, Room MC 3-559
1818 H Street NW
Washington D.C. 20433

For "Worlds Apart" see


For papers see also:

             Andrea Bennett                                                     
             m>                                                              To 
             Sent by:                   
             owner-statalist@hsp                                             cc 
                                         Re: st: RE: Your opinion on income     
             06/07/2008 02:01 PM         groups and inflation                   
              Please respond to                                                 

Thank you both for you suggestions! So, I will use NO dummies for
income since there is no/I have no theoretical expectation that the
effects should be different within each specific level of income,
rather I do expect one direction of the effect of income on y, only. I
might test for an interaction dummy with the highest income class,
though. In the case of the "trust" variable I will NOT use dummies,
too (for the above reason).

Many thanks for you little helper!

On Jun 7, 2008, at 7:40 PM, SamL wrote:

> Making ordered categories dummy variables vs. making them one
> ordered variable by assigning values are both okay.  The dummy
> variable approach is most flexible, allowing the effect to be
> different across categories. But, if you have no theoretical
> expectation for that, nor any reason to believe those particular
> categorical lines match the lines you'd draw if you could, then I
> agree with Martin below.
> Sam
> On Sat, 7 Jun 2008, Martin Weiss wrote:
>> The obsession with dummies in some posts is a little bewildering.
>> If you
>> already have an ordered variable recording trust in government it
>> is hard to
>> see why you would want to throw away this information and let the
>> resulting
>> dummies eat up your df. Just include the variable in the regression
>> as it
>> is. If you absolutely want to, -tabulate, generate- could be
>> helpful. Be
>> sure to omit one dummy in the regression to dodge the famous dummy
>> trap...
>> Martin Weiss
>> _________________________________________________________________
>> Diplom-Kaufmann Martin Weiss
>> Mohlstrasse 36
>> Room 415
>> 72074 Tuebingen
>> Germany
>> Fon: 0049-7071-2978184
>> Home:
>> Publications:
>> SSRN:
>> -----Original Message-----
>> From:
>> [] On Behalf Of Andrea
>> Bennett
>> Sent: Saturday, June 07, 2008 7:19 PM
>> To:
>> Subject: st: Your opinion on income groups and inflation
>> Dear Statalisters,
>> I wonder what you do think about the following issues:
>> 1.
>> I have a cross-sectional panel data set with a time span of 10 years
>> (5 complete surveys). The survey does NOT ask for (more or less)
>> precise income but instead for income groups (e.g. between 3000 and
>> 4000 $). As this does cluster the income quite a bit I was wondering
>> of how much use it would be to control for the inflation rate (which
>> was rather low, about 1-2% p.a.)? Still, I tend to control for
>> inflation since there is a gradual shift from lower income groups to
>> higher income groups. Would you include the inflation adjustments or
>> not?
>> 2.
>> Do I need to treat these income groups as categorical data and
>> therefore generate dummies for each income category? Do I also need
>> to
>> control for the highest income group with an additional dummy since
>> it
>> measures income>=10'000$? This just affects the df quite a bit, I
>> think. How would you deal with such a variable?
>> 3.
>> Related to the above question, I wonder how one usually deals with
>> interval variables. I have a variable measuring the trust/confidence
>> in the national government on a scale of 0-10 (1-point steps). Again,
>> as I have learned it, I would have to include a dummy for each trust-
>> level. Again, losing df is what brings me to ask you again. Do I need
>> to generate dummies here or not?
>> Kind regards and many thanks for your consideration,
>> Andrea
>> *
>> *   For searches and help try:
>> *
>> *
>> *
>> *
>> *   For searches and help try:
>> *
>> *
>> *
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index