I would recommend using microsimulation techniques in this particular case, especially seeing that you have auxiliary information until 2015. Depending on the technique, this comes close to MI. However, alternative approaches might be more useful. If you are interested, you can contact me directly for some introductory literature.

I would recommend not to impute the data, or at least present results with and without the imputation. If the results change a lot, I would not be confident about the imputation, which posses a question about why would you make an imputation in the first place. 

Hi David,

Thanks for your contribution but how defensible is using a data (from a very credible source) which ended at 2011 to undertake a study in 2016? I'm of the view that if I'm able to impute the data to cover years 2012 to 2015 I can then subject the results to: robustness checks, sensitivity checks and reality checks by comparing results when data from 1980-2011 and 1980-2015 are used....I really don't want to commit econometric blunder, but I know that MI is often used a lot...I'm just saying.

Hi Ngozi, 

I guess it depends on the nature of your analysis, what you are trying to achieve with your research. If you are trying to predict Gini, of course it is justified. If you are trying to provide a reasonable explanation of changes in inequality in the last 20 years, imputation seems to be less reasonable. It will also depend on how much of the Gini are you able to predict during the period you have data. 

It is true that data up to 2011 may be a little old, but if that is what is available there is not much to do (BTW, did you check POVCALNET?, maybe there is more recent information  there).  One option is to present the paper with data up to 2011, and then a section about "Further developments" or something like it where you use the data that you have up to 2015. You could state something like: "the results of the previous section and the evolution of my explanatory variables for the period 2012-2015 suggest that ..." 

This is my opinion. As a reader, if I see both calculations 1980-2011, 1980-2015, and they are different, I will undoubtedly pay attention to only the first one. In which case imputation will only be defensible (for me) if it does not impact in the results. However, an extra section where you tried to use your results to infer recent patterns, sound interesting and more defensible, and if a reader does not like the imputation she could just skip that part.  

Good Luck!

David,

I'll go with your point of view. I'm also trying to tread with caution so that my results are at best correct and defensible. I' ll apply MI for the missing years and undertake comparative analysis of my results.

I also accessed POVCALNET via this link http://iresearch.worldbank.org/PovcalNet/povOnDemand.aspx but couldn't get much from there. I only got the default poverty line for the respective country....and then I remembered that Frederick Solt used the POVCALNET data for his Gini calculations.

Thanks again, David....your comments are quite apt!

....gracias!!!

Select a country/es there, then when you go to the detail output table for each year you have the Gini (you have to click in each year). However, it seems out of data for SSA countries. 

...exactly!. I'll stick to Solt (2009, 2014) and take it from there...

As I understand your problem, you postulate that Yt = f(Xt) + et where Y is the dependent variable, F(X) is some function of independent variables, and e is well-behaved error term that is likely correlated over time. You always observe X but you do not observe Y when t > T. The strong assumption is that the function that is estimated when t T. I do not see how to test that assumption, but on the other hand, the assumption is not outrageous and most journal reviewers would likely accept it provided you justify the stability of your model over the observation period.

I do not see this as a multiple imputation problem so much as I see it as a time-series projection. I like the idea of a micro-simulation. I believe I would use Bayes estimation to perform that micro-simulation.

Like David, I'd be cautious about imputing the extra years.   

1.  You are much shakier ground imputing beyond the time period for which we have data then you are imputing between existing datapoints.

2.  Remember that SWIID provides imputed data.  So you would be estimating inequality (using MI) based on estimates of inequality (created with MI) based on estimates of inequality (based on surveys).

3.  Take a look at: Jenkins, S. P. (2015). World Income Inequality Databases: an assessment of WIID and SWIID. The Journal of Economic Inequality, 13(4), 629-671. (pdf available here: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.688.8753&rep=rep1&type=pdf)

4. I think the most recent World Bank Data is here: http://wdi.worldbank.org/table/2.9 and here: http://databank.worldbank.org/data/reports.aspx?Code=SI.POV.GINI&id=af3ce82b&report_name=Popular_indicators&populartype=series&ispopular=y

Hi William,

I appreciate your comments. I also agree that micro-simulation may well address this problem better than the MI or MID (multiple imputation, then deletion) approach since I'm considering imputing future data (i.e. 2012-2015), more so that my data is TS in nature. 

...and thanks a lot Daniel for the link to Jenkins (2015) paper which I have downloaded already. Also, the World Bank Table 2.9 has 45 SSA countries with 17 countries having Gini indices from 2011 and above. I'll attempt micro-simulation and see how it goes...

Thanks.

Hello do yo solve this issue? as I have similar issue, maybe you want to hear about my solution

Hi Antonio R Andres kindly share your solution.

Thanks!

Dear Ngozi and Antonio, I am not sure if you have checked Lahoti et al. (2016). I suggest you see their Global Consumption and Income Project (GCIP) database. They have an extensive dataset on income inequality and poverty spanning 1960 - 2015/16.

It might be helpful.

Article The Global Consumption and Income Project (GCIP): An overview

http://gcip.info/

https://www.econstor.eu/bitstream/10419/142119/1/860670759.pdf

Thank you

Thanks for your contribution Isaac K. Ofori

I have been using the Lahoti et al (2016) GCIP dataset since 2016...highly recommended to anyone working on inequality and poverty.

Hello Ngozi and Antonio

I am facing the same problem. Have you found a solution please ?