Help with stats work

#1
Hi everyone at Talk Stats,

I've really been struggling on the correct method to use to prove/disprove if my results are statistically significant. For the context in 2006/07, the UK tuition fees have changed from £1k to £3k and then again in 2012 to £9k. Attached is a text file of my collected data, it contains all the UK higher education institutions from 2000/01 till 2016/17 I need to show if there have been any statistically significant changes in total income and total expenditure in relation to these key dates. I have access to SPSS and STATA but more importantly, I'm not looking for an answer I want to understand how I can get this to work

Thank you all for your help
Dan
 

Attachments

obh

Active Member
#2
Hi Dan,

When you use sample data and you try to get conclusions from the sample to the population, then you may say that the results are significant if the behavior of the samples represents the populations.

I have a feeling that you have the population data?

or do you want to check if the change on the fee influences the income?
 
Last edited:
#3
Hi Obh,

I would like to see if these changes in fees do influence income and expenditure. income isn't really that important, Higher Education institutions (HEI) don't have much control over their income, their expenditure should be more interesting.

My data is a list of every university in the UK between 2000/01 to 2016/17 all expenditure and all income. I have breakdowns of how these are split but that isn't so important just yet. I'm more interested in the big picture first, then I'll look at how the changes have impacted the expenditure profile splits.

Data looks like this

1562599363365.png

Think you could help?

Thanks
Dan
 

Miner

TS Contributor
#4
Trying to do what you want on the entire data set will be problematic because of the variation between institutions. i would recommend that you analyze select institutions that represent small, medium and large institutions individually using time series analysis. You can de-trend the data and look for discontinuities at those time periods. I recommend that you regress expenditures on income. I think you will find that they spend their entire income.
 

noetsi

Fortran must die
#5
To address minor's point, you can try a multi level regression where expenditures and the change in fees is nested inside school (or perhaps size of schools as he suggests). The DV would be expenditures and changes in fees a predictor. If p is <.05 it suggests that change in fees influence expenditures although you really need to control for other variables I am sure.
 
#6
Miner, I have the university rankings I could do the top 10, middle 10 and the bottom 10? also what does de-trend mean? it's actually interesting before 2008 income almost matched expenditure, but since then the income is consistently about 5% higher than expenditure

Noetsi, what could I use as a control? I have every institution from the uk in the data, I suppose the top universities would be almost completely unaffected by any changes in fees but that's quite an assumption, I don't think i could look outside the UK either

Thank you
 

Miner

TS Contributor
#7
Miner, I have the university rankings I could do the top 10, middle 10 and the bottom 10? also what does de-trend mean? it's actually interesting before 2008 income almost matched expenditure, but since then the income is consistently about 5% higher than expenditure
De-trending subtracts the linear trend component from the data making it easier to see a potential discontinuity or jump in the data.

Cambridge expenditures:
Cambridge.png

This is an Individual control chart used in industrial statistics. It tests whether a process that generates times series output is in a state of statistical control, that is it is stable and predictable. This chart of the de-trended data tells me that there were no abnormal deviations (i.e., changes resulting from tuition increases) and that left alone the expenditures will continue to increase along the linear trend line.
I Chart of DETR1.png