logit probit difficulties

[tona]

I am relatively new to eviews and finding out that I am also very unskilled. I am currently doing an econometric project on how income determines happiness. The data I have collected presents happiness in binary form. Therefore my dependant variable is binary (0 or 1, unhappy or happy respectively) and I subsequently have to utilise probit and logit estimation models. I have been advised to use lpm probit and logit techniques and to compare them for results. I know that the coefficients created by probit and logit estimation models are not completely representative and I have to calculate the marginal effects which I think I can do. I do have some general questions that I could really do with some help on: :)

1. Many of my variables that I have used in the model appear insignificant with a p-value of greater than 10% including income. Does this mean that, for example with the coefficient of income at 1.4 (larger than any other variable) (when running LPM) that I cannot say income has a larger impact on happiness than the other variables yet there is lack of significance. OR because of the lack of significance I can not mention the size of the coefficient at all? (If I have made any sense there)

2. I am unsure when to take the natural log of variables. I have been taking the 'ln' of variables such as: income, age, traveltime, workhours, amountsaved. Is this correct or at all relevant?

3. If I wish to add quadratic variables for the likes of age and income. Is it it okay that I square the natural log of the variable: for instance (lnincome)^2? If that is correct and I receive 0.12 coefficient for lnincome and -0.08 for (lnincome)^2 is it possible to calculate the maxima for this variable? if so how?

4. I understand that there are many diagnostic tests that can be run to test your model when the dependent variable is continuous, however I understand that there are less or little that can be done when using logit and probit with a binary dependent variable. What diagnostics can I run to test my model?

5. When variables are insignificant would it be wise to leave them out of the model completely or is it likely that there are underlying problems with my selection of variables?

6. I have a variable called 'incomediff' where I have calculated the difference of every individuals income from the sample average. When I do not include this in my model the coefficient for lnincome is 0.34... however when I include incomediff the lnincome coefficient jumps dramatically 1.84. What the hell is going on? The p value for incomediff when it is included is 0.11. It may be also important to note that without incomediff in the model lnincome's p value is 0.35 however with incomediff it lowers to 0.13. What should I take from this?

I am terribly sorry for such a long post but I really need your help. I am also sorry for my incredulous lack of knowledge.

Thanks, any help is more than welcome :D