Panel Regression Step-by-step Procedure

[harata25]

Hi All,
I am not very good in econometrics and eviews, but i am trying my best and need help here.
I am testing determinants of entrepreneurship, using panel data of 7 countries with 10 years annual data for each country, so that's a total of 70 observations.
I have about 11 variables (i know this is a bit much, but I will eliminate them later. I start off with many variables, and eliminate one by one the insignificant ones).

Next, I have stacked them up on excel and created a workfile, imported them and all is fine now. All in raw data. Some of the independent are in normal units (eg million of population) while some are in percent (eg % of GDP). Do I need to take log for all the variables? Or just those with normal units, and leave the percent units variables as they are?

Next, which option is right and which should I do?
Option 1 - I estimate the model first, then eliminate the insignificant ones, and I am left with about 7 significant variables. Then, check for fixed/random effect and decide whether I use fixed or random.
Option 2 - I estimate with all variables included and test for whether I should use fixed/random. Once I got the answer, I start removing the insignificant variables.
Which is correct, option 1 or 2? Because with option 1, once i remove the insignificant variables, as soon as i change to fixed effect all the significant variables now change their p-values again and most become insignificant again. Please help.

Next, what should I do after this? Check for unit root using the Im, Pesaran, Shin and the Levin, Lin, Chu test? Or what should I do next?

Please kindly advise. I am really stuck and need a step-by-step procedure on running this whole thing.

Many many thanks guys!