Hi everyone! I would like to ask your opinion about some transformations that i need to do in SPSS I am doing a regression analyse on the impact of schooling on wages. I’m using data from the European Social Survey (ESS). My problem concerns defining the wages of the respondent. Let me explain: ESS provides information about the net income of the household but of course doesn’t give exact numbers. It only shows in which salary-interval the respondent is situated. For my work I would like to do some basic descriptive analysis of the wages but I don’t know how I should best tackle the job. Anyone any suggestion? Thanks!!
Note: I was thinking to take the middle class of each salary-interval and use that to calculate the means and stddev. But the last salary-interval is “6000€ a month or more which makes it impossible to take the middle class. Also: the size of the intervals aren’t the same.
Do you have access to the distribution (i.e., from a large sample of govt stats) for each country? You can use this (coupled with some assumptions) to get your estimates.
Thanks for answering Daniel Wright! The data to which i have access is the following:
- Every respondent who took answered the whole survey (no missing values) was asked in which decile his/her income was situated (1st, 2nd,.., 10th). The survey's appendix gives more information about the minimum and maximum amount of each decile (i.e. decile 3 is a salary between 3940 and 4920).
My mentor asked me to find (based on the survey data) a way to estimate the average wage for each country and to use the "recoded" data concerning the deciles to perform the regression analyse.
(Note: I'm doing a comparative analyse of 21 countries.)
Mohammed Ouamari, if I follow, you are asking how to compute the mean for grouped data using SPSS. Is that right? If so, I think the AGGREGATE command is your friend. See if you can adapt relevant portions of the following example to do what you want. (The first part of the code is just generating a data set that resembles yours. For you, the interesting bits are in Part 2 of the code.) HTH.
* Part 1: Generate some data resembling
* the situation Mohammed has described.
* I'll use Age rather than salary as
* the variable of interest.
NEW FILE.
DATASET CLOSE ALL.
* Modify path to show where you store
* the sample datasets for SPSS.
GET FILE='C:\SPSSdata\survey_sample.sav'
/KEEP=id age.
DATASET NAME raw.
* Suppose age had been collected as age decile.
* Generate those deciles to illustrate.
RANK VARIABLES=age (A)
/RANK
/NTILES(10)
/PRINT=YES
/TIES=MEAN.
RENAME VARIABLES Nage = AgeDecile.
MEANS age BY AgeDecile /CELLS=MIN MAX MEAN COUNT.
* Now imagine that we do not have the original age variable.
* We have only the age decile for each person, and the
* min and max values for each decile.
* Let's use AGGREATE to create a new dataset holding
* all that info.
DATASET ACTIVATE raw.
DATASET DECLARE deciles.
AGGREGATE
/OUTFILE='deciles'
/BREAK=AgeDecile
/age_min=MIN(age)
/age_max=MAX(age)
/N_BREAK=N.
* The newly created 'deciles' dataset
* contains Min, Max and N for each
* of the age deciles. Use that info
* to estimate the mean for Age.
* Part 2: As Mohammed already has his deciles, this
* is where he will want to start adapting code.
DATASET ACTIVATE deciles.
SELECT IF NOT MISSING(AgeDecile).
LIST.
* Using Mean(min,max) as the estimated mean for
* each decile, estimate the grand sum of all ages,
* and divide by the total N to estimate the overall mean.
* Use Mean(min,max) as the midpoint for each interval.
COMPUTE mid = MEAN(age_min,age_max).
* Compute the estimated sum of the ages for each group.
COMPUTE AgeSum = mid * N_BREAK.
* Now get the sum of the sums and the total N.
AGGREGATE
/OUTFILE=* MODE=ADDVARIABLES OVERWRITE=YES
/BREAK=
/N_total=SUM(N_BREAK)
/GrandSum=SUM(AgeSum).
* Compute the estimated mean for Age.
COMPUTE EstMean = GrandSum / N_total.
VARIABLE LABELS
EstMean "Mean estimated using mid-points of age deciles".
DESCRIPTIVES EstMean.
* Compare this estimate to the mean of the original age variable.
DATASET ACTIVATE raw.
DESCRIPTIVES age.
* End of code.
For those who do not have SPSS, the results show this:
- Mean estimated using mid-points of deciles: 45.76
- Mean age using the original raw data: 45.56
So for this example, estimate using mid-points of deciles is not far off.
To be clear: the salary are on the same level as the education.
Maybe you should try to use Ordinal Regression Analysis instead of using Interval Regression Analysis. You would not need to estimate the mean for each respondent. Do you see large differences between countries for the highest categories?
Maybe you might need a multilevel ordinal regression analysis to have a better analysis for your data. Maybe use auxiliary information on income distributions in each of the countries (maybe from the European Statistical Bureau). These analyses can be performed in most software available (R, MLwin, Statistica, SPSS or SAS) Due to different algorithms the results can differ between software.
I assume you do NOT want the midpoint of the interval because that assumes values are distributed such that this is the mean (so, it would be fine if the distribution were uniform). With salary data you will have bumps and spikes and odd values around tax thresholds etc., and these will vary by country. You send "mentor asked you." If this is an assignment, you probably just want to explain the issues to her or him. If this is for a publication it will require some work.
You may convert the categories to ordinal numbers, such 1,2,3,... starting from the lowest to the highest. Then, replace them back when you're writing your manuscript, or use them and explain that in your methods sections. Best of luck!
If you insist to convert it to numerical values, look for the "German Tank Problem".
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude