Problem with small amount of cases

[Cai]

Hi

New here, and fairly new with statistics.

I have a problem with too few cases, but more than enough variables and observations. I have no background in statstics, but chose to use it as method in my dissertation because I wanted to learn new stuff.

I have annual quantitative data from comparable and uniform 21 industrial production lines from the same multinational company including profitability (2 indicators 1998-2006), environmental impact (8 indicators 2000-2006), occupational health and safety (3 indicators 2000-2006), and two sets of employee opinion surveys (EOS) conducted in 2001 and 2004. The EOS data is complete for the multinational company data originates from with over 10 000 respondents on around 100 questions each time. I have isolated the people working on the production lines and summarized their answer into one per production line.

The thought is (was?) to run multiple regression to see what connections there might be. But MR needs way less variables and way more cases, according to all standard sources.

What should I do?

 

[CB]

The thought is (was?) to run multiple regression to see what connections there might be.

You kinda need to have a clearer idea of what you're actually looking for than this before you can select an analysis What are you actually trying to find out?

 

[Cai]

What I try to do

What I try to find an answer to is if there is any possible correlation between corporate social & environmental performance (CSEP) and corporate financial performance (CFP), either way. Earlier research is inconclusive though leaning towards CSEP->CFP, and has not had the kind of precise data available that I got my hands on.

 

[bonedoctor]

Need more specifics, sorry

What I try to find an answer to is if there is any possible correlation between corporate social & environmental performance (CSEP) and corporate financial performance (CFP), either way. Earlier research is inconclusive though leaning towards CSEP->CFP, and has not had the kind of precise data available that I got my hands on.

Sorry but I think what CowboyBear was meaning was to be more specific about the kind of data you are dealing with - is it categorical (ordered or otherwise i.e. good/average/bad or north/east/west), numeral (normally distributed or not) etc.

Is CSEP or CFP a monetary unit, a category, how is it defined etc?

The reason we need to know what kind of data it is is that it affects the kind of stats that can be done. One example is in population studies to change a numeral variable (eg age) into an ordered categorical variable (e.g. age bracket 20-30, 30-40) so that instead of an effectively infinite number of possibilities you may have only ten.

Once you've identified any variables which seem promissing this way then correlation might be the way to go but, again, need a bit more info first.

 

[CB]

True - my first thought was "What are you trying to find out?", but of course once this is answered we need to know about the data, so I should've mentioned that too!

 

[Cai]

Data I have

Sorry guys, of course you need more info. Here is intially what I have:

CFP is measured in my case by Ebitda and ROCE, but basic financial concepts using money ( in my case euros) as measure. So metric ratio scale.

CSEP is formed by the following:
Effectivity (1 variable) is also metric, reporting production efficiency in % of ideal circumstances.

Environmental variables (8) are also metric ratio, reporting for example air emissions (SO2 emissions kg per ton product produced) or water emissions (Chemical oxygen demand in kg per ton of product produced).

Occupational health and safety indicators (3) are the following: Absence frequency is calculated as the number of sick-leave days divided by number of sick-leaves in total. Accident frequency is calculated as the number of accidents at work resulting in one or more days lost in sick-leave divided by 1 million. Severity rate is calculated as the number of days lost resulting from accidents divided by the actual number of accidents reported

Employee opinion surveys (2 sets 2001 and 2004) have interval scales where the answering options to the questions were on a scale of 1 to 5, (Disagree, tend to disagree, ?, tend to agree, agree).