# Will Markov Chain work here?

I have a poverty dataset with rows containing different places and different years and together with the poverty percentages in those places. I have to predict the poverty percentage for next year. (Note: 50% poverty means 50% of the population is under the poverty line)

Overall I have over a thousand rows. But I don't have enough data points for each of the places. Actually, the max data point I have for each of the places is only 2.

Example of what the data looks like:

Place    Year Surveyed    Population    Poverty
X        2006             100           50%
X        2012             100           52%


I wish to predict whether Place X will still be in the same state in year 2018. Will Markov Chain work here? If so, what are my states?

• To answer your question of what your states are, I think you need to think about what you mean by state in "will still be in the same state"? Is this the poverty level as a percentage? Is the percentage an integer or continuous? Or is it some categorical variable like "Place X is poor/average/well off". To predict poverty percentage at the next polling cycle I think that a logistic or beta regression would be more direct and suitable. – Alex Mar 1 '17 at 2:05
• Are you saying that if the states are categorical it should be logisitic regression? What if the states are as a percentage (continuous)? – Katherine Mar 1 '17 at 3:17
• If you want to predict whether a place will fall into a category poor/average/wealthy you should probably use ordinal regression. If regressing against fraction of population in poverty I would look at beta regression: jstatsoft.org/article/view/v034i02/v34i02.pdf. If regressing against whether each member of a population in a place is going to be living under the poverty line (i.e. 1 = poverty, 0 = above), then I would use logistic regression. – Alex Mar 1 '17 at 10:52