# Which Statistical Hypothesis Test is appropriate Here?

I have two contingency tables (screenshots below) - This is the results of age range 19-30, I also have two tables for above 30s. I am requesting help with rejecting a null hypothesis.

The data was gathered from a questionnaire. The participants were told what Alexa and Ring devices are, then they were asked whether or not they would buy an Alexa or Ring device. They were then told potential security and privacy flaws in the devices and asked again if they would purchase an Alexa or Ring device (evidently more said no after being told the security and privacy flaws).

I have been told by a friend to use either Chi-Square (with Yates correction) or Fisher's Exact Test to analyse this data to see if my null hypothesis can be rejected. The null hypothesis is: "There is no drop in willingness to purchase devices once being told security flaws" - Obviously there are more people who say "no" after being informed of the flaws, but, is the drop low enough to prove significance?

Is this... right? Is there something else I should use? I'm a total beginner - sorry if this is the wrong place to ask as well.

I was watching some videos, reading some resources and I have been heavily confused. Apparently, I should use McNemar's Test but I simply cannot because there is always going to be 0 "No Yes" answers, there is not a single answer in my survey of "No Yes"... I mean, imagine the scenario where someone says "Yes" after "No":
"No I do not want to purchase the device knowing the benefits" and then "Yes, I now want to purchase the device given I understand its security flaws (of which there are many)"

Thank you for any insight.

Here are the tables for people above the age of 30, too. Just in case you can provide any insight on how to compare the two age ranges!

So, in short... what do I use? Why? Most of my doubt comes from this source: https://libguides.library.kent.edu/spss/chisquare

• Clarification questions: 1) Are you interested in looking if there is a shift in people intending to purchase the device among all your respondent (i.e. running one hypothesis test), or that for each of your four groups (i.e. running four hypothesis tests, testing 19-30/Alexa, 30+/Alexa, 19-30/Ring, 30+/Ring)? 2) Do you have access to the changes to the responses on an individual level (i.e. the "No Yes", "Yes Yes" answers)? If so a sign test may be appropriate. – B.Liu Jun 15 at 14:42
• Your friend's suggestions look as if they would be testing "there is no change in willingness to purchase devices once being told security flaws", and potentially might reject that hypothesis if willingness for both went up, or if one went up and the other down. But it seems fairly likely willingness went down, and you should decide before you test whether you are interested in an Alexa / Ring distinction or an age distinction in the extent to which this happened – Henry Jun 15 at 15:01