How can significance be determined from a "Yes/No" questionnaire? I'm really new into statistics and I need to determine if the change in a dataset is significant.
I collected data from three groups of people (first, third and fifth semester classes from college), where the participants answered whether they use certain substances or not. I need to check if the semester of undergrad has correlation with the use of those substances.
I observed that: 


*

*35% of first semester students use those substances

*8.7% for third semester students

*0% for fifth semester students.


I hope I could make the objective clear since I don't know proper scientific vocabulary. How can I determine significance in this case?
EDIT: I have the datasets in a Google Spreadsheet (from Google Forms) and I do know how to use basic spreadsheet formulae.
 A: Basically you need to check if the yes/no distributions (also called Bernoulli distributions) for the three undergrad semesters are significantly different from one another.
Intuitively speaking you need to check, if the bar plots illustrating the frequencies of occurrence of yess and nos for the three cases are different enough given the sample size you have. If your survey data is extensive and the three bar plots still "look" very different, then you can be quite certain to have a correlation between substance use and semester. Conversely, if the true (but unknown) distributions are markedly different from one another, but your sample size is small, then you may still not rule out the  null hypothesis of both being equal.
Now, on the more rigorous side there are a plethora of methods to check for such type of significance, with the most prominent one possibly being the (two-sample) chi-squared test
https://www.itl.nist.gov/div898/software/dataplot/refman1/auxillar/chi2samp.htm
As in your case however you have three such tests (1st vs. 3rd, 1st vs. 5th and 3rd vs. 5th semester) you need to adjust your significance level accordingly, taking care of what is called the multiple comparisons problem. The easiest way to solve this and that should work here is the Bonferroni correction, which involves dividing the chosen significance level by the number of tests you make. So, for example, when you have agreed on a significance level of 5% for a single test, then you should rather use $5/3$% for each of the three tests you make.
So, in a nutshell, here's my suggestion:


*

*Go for the (two-sample) chi-squared test

*choose a significance level $\alpha$ as you would do for any other hypothesis test

*perform a chi-squared test for the three combinations of semesters each, i.e. 1st vs. 3rd, 1st vs. 5th and 3rd vs. 5th, using $\alpha/3$ as significance level for each test

*If the result of ANY of the three tests is significant, then you may infer a statistically significant dependence between substance use and semester


Hope this helps...
