Jolene, although you could use a chi-square, my preference would be to use logistic regression with your 3-point scale as the predictor and the dichotomous variable as the outcome.  The resulting odds ratio will be much more informative about the behavior of your 3-point scale than the typical chi-square output would provide.

And if you are willing to move to R (https://www.r-project.org/), you can start looking at Binomial or Poisson GLMs (http://www.statmethods.net/advstats/glm.html) and much more depending on your data, distributions... 

.

.

chi2 test will only allow you to test (therefore possibly reject at a given confidence level) the independence between your two variables

if you reject independence, you still do not know why and have to go for a post-hoc analysis

http://pareonline.net/getvn.asp?v=20&n=8

.

20 samples on a 2x3 contingency table leaves very few samples per cell (if any) ... you might be better off with the Fisher exact test !

https://en.wikipedia.org/wiki/Fisher's_exact_test

and for the post-hoc test ... well, i hardly ever went beyond eye-balling (shame on me !) or comparing residuals

.

(Alexander and Michael suggest faster ways ; however, at the outset of such analysis, you still have to test the significance of the results)

.