Sanja -

That does not sound like much data with which one might work.  For example, if you want to look at medium size companies as a stratum or as a subpopulation for which you want to estimate means or totals or proportions, then a sample size of one is completely inadequate, no matter how small the variance, particularly since it is impossible to even estimate a variance from a sample size of one, i.e., n=1.  I assume that this is a sample, and that the subpopulation of medium size companies is not N=1, but N>1.  However, if this is a census, then all you need are the usual descriptive statistics. 

You might be able to use regression to find relationships between the seven variables, for these 12 companies, but again, you can only learn about the companies in your sample.  You cannot make inferences to a larger population. 

However, assuming this is a sample, perhaps you have a continuous variable which can be used as a measure of size, and you have that size measure for every member of your population (or at least for each member of the sample, and a total for that size measure variable for the remainder of the population, not in the sample).   In that case, one might use regression to provide inferences.  You note different sizes of companies.  How did you measure that?  If you have that measure for each member of the population, as a continuous variable, then you can use regression to 'predict' (an estimate for a random variable) totals for variables related to that size variable.  If the relationship is linear through the origin, and the classical ratio estimator has the appropriate heteroscedasticity, then even the variance of the error of the predicted total can be estimated with only the total of the size variable for the population, and the individual contributions associated with each member of the sample. 

If you do have a linear relationship through the origin between a continuous size measure variable and a variable of interest, and you have the size for the total population, you might be interested in the following:

 https://www.researchgate.net/publication/319434665_Comparison_of_Model-Based_to_Design-Based_Ratio_Estimators

or perhaps even this:

 https://www.researchgate.net/publication/319914742_Quasi-Cutoff_Sampling_and_the_Classical_Ratio_Estimator_-_Application_to_Establishment_Surveys_for_Official_Statistics_at_the_US_Energy_Information_Administration_-_Historical_Development

Cheers - Jim

Dear James R Knaub,

I really thank you for such detailed reply which will be very useful in my future work.

Best regards,

Sanja

It seems very small sample size relative to the number of variables, and it is a great advice from Prof James R Knaub James R Knaub

Dear Ayman Amin,

thank you very much for the advice.

Best,

Sanja

No. of sample is important here.Pl mention that in question.

Dear Ajit kumar Roy,

thank you for the reply.

Best,

Sanja

Don't rely on parametrics (mean, std deviation, t test, Pearson, ANOVA) unless variables pass tests for normality (e.g., Shapiro-Wilk). Use Spearman coefficients, Mann-Whitney, Fisher Exact, Friedman.

Dear T.L. Brink,

thank you very much for the answer which I will follow.

Best regards,

Sanja

First of all, the sample size should be adequate enough to achieve accuracy of whatever the analysis will be done.

Second, the normality of the data should be assessed to determine which type of test to do (parametric/non-parametric).

Third, it should be made clear whether the companies will only be compared based on the 7 variables or any association among the variables is to be established.

Dear Dipayan Datta,

than you very much for the useful reply.

Best regards,