Valentino Dardanoni, Mario Fiorini and Antonio Forcina, "Stochastic
Monotonicity in Intergenerational Mobility Tables", Journal of Applied
Econometrics, Vol. 27, No. 1, 2012, pp. 85-107.

The first part of the paper (unconditional analysis) uses data from

Ganzeboom, H. B. G., Luijkx, R. and Treiman, D. J. (1989),
"Intergenerational class mobility in comparative perspective", Research in
Social Stratification and Mobility, 8:3, pp. 3-79,

which may be found in the file treiman-data.txt, which is zipped in the file
treiman-data.zip. This is an ASCII file in DOS format; Unix/Linux users
should use "unzip -a".

The second section of the paper (conditional analysis) uses data from the
National Child Development Study. The NCDS is a continuing longitudinal
study that seeks to follow the lives of all those living in Great Britain
who were born in one particular week in 1958. Users registered with the
Economic and Social Data Service (ESDS) have access to the NCDS datasets via
the instant download service or can analyse, visualise, subset and download
selected data from NCDS via the online Nesstar software tool:

http://www.esds.ac.uk/longitudinal/access/ncds/l33004.asp

This paper makes use of Sweeps 0-5 as well as the NCDS exams file (1978).
The original data files have been merged and coded in Stata. The do files
are available upon request. 

Two data sets are provided in ASCII format in the files dfullm.csv and
dfullm_inc.csv. They are zipped in the file NCDS-data.zip.

These dataset are also provided in binary, Matlab files called dfullm.mat
and dfullm_inc.mat. Along with two temporary Matlab files, dnew.mat and
temp.mat, they are zipped in the file NCDS-data-matlab.zip. The "-a" option
of unzip should not be used with this file.

These two data files include 13 variables. There are 1942 observations in
dfullm and 1104 observations in dfullm_inc.
We select individuals who:

* are males
* have a natural father 
* have cognitive skills scores
* have non-cognitive skills scores
* have reported education qualifications
* have reported father's age 

Moreover, in dfullm we select individuals for whom we observe father and son
social class, while in dfullm_inc we select individuals for whom we observe
father and son wages. 
                                    
Each of the dfullm and dfullm_inc files include 13 variables:

 1. Son social (dfullm) or income (dfullm_inc) class. Coded from 1 (lowest)
    to 6 (highest). 
 2. Father social (dfullm) or income (dfullm_inc) class. Coded from 1 (lowest)
    to 6 (highest). 
 3. O Level dummy. 
 4. A Level dummy.
 5. Higher Education dummy. 
 6. Principal component cognitive skills (age 7).
 7. Principal component cognitive skills (age 11).
 8. Principal component cognitive skills (age 16).
 9. Non-cognitive skills factor #1 (age 7).
10. Non-cognitive skills factor #2 (age 7).
11. Non-cognitive skills factor #1 (age 11).
12. Non-cognitive skills factor #2 (age 11).
13. Father's age in 1974.

Note: This order should be kept in mind when selecting the control variables
in the son and father's marginal and in the odds ratios. The order is also
important to read the vector of estimated coefficients and standard errors.
Note that, in the final tables of the paper, where the betas and z ratios
are reported, Father Age is the first covariate reported, but it is actually
the last variable in the estimated beta vector from Matlab.  

The zip file dff-matlab.zip included all the matlab programs needed to
perform all the analyses described in the paper, as explained in the file
dff-user-manual.pdf. The Matlab files are ASCII files in DOS format.