# Large (n > 1,000) Interval Censored Datasets

Is anyone aware of fairly large publicly available interval censored datasets with covariates? I would like to use one for demonstrative purposes for an R-package.

One that comes to mind is the tooth study, (see 'tooth24' in the R-package 'straweib'). However, despite being in several R-packages, this dataset is part of a private study and isn't available for use without permission. Also, it'd be nice to have new datasets in the statistical literature.

You can have a look at the European Social Survey. In this survey, household income is an interval censored variable. The survey also provides plenty of variables that can be used as covariates. For the reference year 2012, the complete survey comprises more than 54 000 individuals from 29 countries. The sample sizes for individual countries vary between 752 for Iceland and 2958 for Germany.

The survey documentation has an appendix on the definition of the income intervals.

For a given country the income intervals are disjoint. However, each country has its own definition of the income intervals. That means if you merge the data of two or more countries, the income intervals become overlapping.

The data can be used "without restrictions, for not-for-profit purposes", as long as you cite it properly (cf. European Social Survey (ESS) - Conditions of use).

• Quick question; when you say income is interval censored, do you mean disjoint intervals? My first guess is that the interval censoring would be disjoint, i.e. [0, 10k) [10k, 20k), etc. What would be a little disappointing to me is if there were no overlapping intervals (i.e. subject 1 = [10k, 50k), subject 2 = [25k, 100k), etc.). The reason I care about overlap is that I'm working on an algorithm for a semi-parametric model. In the case of no overlap, you don't actually need a special algorithm, you can get it from the standard Cox-PH/KM estimator with some very mild data manipulation. May 27, 2015 at 16:41
• I have updated my answer. Your first guess is correct, but not entirely ... May 27, 2015 at 18:51
• Ah, excellent. Very much appreciated! May 27, 2015 at 19:24
• Quick question: is your handle your real name? I'm getting ready to submit a paper that used this dataset, and I'd like to at least mention thanks in the paper. Jun 19, 2015 at 2:37
• Rather than my name, you should cite this website in your thanks. Jun 20, 2015 at 15:09