OLS regression - robust estimates for parameter's variance I'm estimating a model for corporate social responsibility (not important). 
I have found my variable of interest significant at 5% confidence level. 
My sample is $N=84$, cross-section.
For this I used OLS estimation.
Then I doubted if I should have estimated the model with robust estimators (that is, HAC, HC1, ...), and did it. It turned out that the parameter is now only weakly significant (i.e. not at 5%, but at 10%).
My questions are: 


*

*with such a small sample $N=84$, is it appropriate to use robust estimators? 

*can I test for heteroskedasticity some other way, so I can rely on my non-robust estimation?

 A: Moving my comment to an answer, first test whether you need a heteroscedasticity correction at all. As we discussed the HAC standard errors don't make sense in the context of cross sectional data but I now read up on the HC0, HC1, HC2, and HC3(a) standard error corrections and those are indeed intended for correcting heteroscedasticity. HC0 is for estimating Huber-White standard errors.
However, if you cannot reject homoscedasticity then robust standard errors will not improve your inference. In fact, given your small sample size the robust standard errors can be biased even if you have heteroscedasticity in your data. See Imbens and Kolesar (2012) for more information on this and some very useful practical advice on using robust standard errors.
If you find heteroscedasticity and you feel that you absolutely need to correct your standard errors for it, have a look at Cribari-Neto and Galvao (2003) or a more recent paper, Cribari-Neto and da Silva (2011). Both papers describe ways to obtain heteroscedasticity robust standard errors in small samples. Given the purpose of your work it may or may not be worthwhile to implement these procedures because both papers are quite technical.
