Recommendations for algorithm

I have a very large dataset consisting of approximately 15,000 rows of survey data. The surveys are quantitative summaries of interviews with investors regarding their rankings of financial asset managers. They rate a manager on a scale from 1-5 and we have approximately 5 years of data for each investor on asset manager.

The data is indexed as:

Investor name | Manager name | Year | Rating parameter 1 | R. parameter 2 | ... | R. parameter 10 |

To give you an idea, I made an example below. Currently the data is saved in an Access-database with a ton of SQL JOINT's - but I can convert the data pretty much how I want to. So compatibility is not an issue.

My question is: I have a wild idea that somehow the answers are correlated between years. So maybe an increase in parameter 1 and parameter 2 is correlated with a fall in parameter 10.

I'm looking for ways to test my hypothesis without having to do it manually (this would be quite tedious). Maybe through machine-learning or subgroup discovery (I have not idea if this is correct).

Are any of you familiar with tools or algorithms that can help me test my hypotheses? Naturally I'll have to make sure my analyses are statistically significant.

I'm looking very much forward hearing what you think.

Example of data

Buffet Holdings | Goldman Sachs AM | 2014 |1|5|5|1|4|5|1|1|1|5|
Buffet Holdings | Goldman Sachs AM | 2015 | 4 | ........| 1 |
Gates Ltd. | JP Morgan AM | 2014 | 4 | ........| 2 |
Gates Ltd. | JP Morgan AM | 2015 | 4 | ........| 3 |
Buffet Holdings | JP Morgan AM | 2014 | 3 | ........| 5 |
Buffet Holdings | JP Morgan AM | 2015 | 2 | ........| 1 |

1 Answer

A correlation matrix would help discover any pair-wise correlation (between any 2 variables). I think that would be a great place to start, as it is just one line of code for initial analysis.

After that, if you want to investigate "an increase in parameter 1 and parameter 2 is correlated with a fall in parameter 10", you could write code to form interaction variables like sum / product / divide and build a correlation matrix between the interaction variables vs. the original variables.

Personally, I found the corrplot R package useful in visualizing the correlation matrix. Hope this helps.