IV and Heckman help for Cross Sectional Data (tight deadline)

[Jenai Rawls]

Using data on sleep in 1975. I have variables earns74(yearly earning in 1974), hrwage(current hourly wage), age, educ, totwork (total minutes worked per week), and yngkid (dummy variable if they have young children). My dependent variable is sleep (minutes slept per wek). Im actually looking for a casual relationship so my variables of interest are hrwage, and possibly earns74.
I am trying to run two separate IV test on my regression, but I have cross sectional data and I'm having some issues figuring it out. I would also like to run a heckman selection for my hrwage variable and i don't know which i should run first (IV or Heckman). I think i should run an IV first in order to solve my endogeneity issues, then run a heckman to solve for the limitations in my data for hrwage.

IV: Could someone help me out with a code or in the very least explain how i know that the IV is goof based on the actual numbers that i would get in stata. The more i research the more confused i get. I should note that my IV do not have to be perfect i just have to run some methods and see how it changes my regression results. I want to test union(which is a dummy variable... not sure if that can be used as an IV) as well as experience and spousal wage. What kind of numbers am i looking for

Heckman: since i plan to use it on a independent variable as opposed to dependent, i would make my dummy variable in the first stage a dependent variable. I was told that i need to have another set of info on the individuals (not just the wages, since you don't know the nature of missing values) but im not sure what he meant exactly.

Any assistance is appreciated.

[Jenai Rawls]

Update:
so i decided to change my IV to years married to get a better result. i first ran education to test the methods i have found and my main variable ended up being omitted due to collinearity so i dont know if thats a bust or its supposed to happen (like i said whether the IV is efficient or not isnt the point of the exercise. its to see how it affects the beta of interest and understanding whats going on and if i would/should keep it in my regression). i never expected experience to work honestly, i expected it to be correlated but i didn't think it would render my endogenous variable useless. Is that something that should happen when its invalid or was it just a bad move.

Code sequence i used:
reg slpnaps hrwage learns74 totwrk yngkid age educ
eststo ols

ivreg slpnaps (hrwage=exper) learns74 totwrk yngkid age educ (here hrwage was omitted due to collinearity)
eststo iv

reg hrwage exper learns74 totwrk yngkid age educ (here age was omitted due to collinearity)
predict hrwagehat

reg slpnaps hrwagehat learns74 totwrk yngkid age educ (here is where hrwagehat was omiited)
eststo twostage

*comparison table
estout

I'm also not sure how to test the correlation of education and my error term. I tested the correlation between education and hrly wage and it was kind of week (i think) with 0.2614



Also, i know i said i had a strict deadline but even if its months later i would love to hear your suggestions. Ill be using these methods for other analysis and even if i don't pass the requirements for this, understanding it down that line will be extremely helpful.

[Jenai Rawls]

UPDATE:
Okay so i have finally realized that my educ and exper were correlated which is why I wag getting those results. I ran it without educ and got results. My question now is do i keep education out of my actual regression or leave it in since im only worried about a causal relationship and concerned only with my beta of interest. tying to figure out what equation i should have when i start my heckman