It's been operationalized in ordinal level. Any attempt to make it appears to be continuous again would require a lot of unrealistic assumptions; I will advise against doing so.
Another reason not to add them up is that the two items concern only frequency but not quantity. Since a fast food meal is likely to be more caloric than a common snack, it's misleading to add them up. Because in a way, we are saying having 6 fast food meals is as bad as having 2 fast food meals + 4 snacks, which is quite far from the truth. Not to mention that though it may be safe to assume most fast food meals are "unhealthy," there is definitely a lot more grey area among "snacks," which can range from deep-fried Oreo to celery sticks.
What to do? I'd suggest don't be overly creative with questionnaire items unless the analysis manual accompanying the dietary measurement specifically describes this kind of additional analysis. Otherwise you'd risk ruining the validity of the tool. Instead, just report the two separately. And if you're going to use them in a regression model, then model them as two set of variables rather than a combination. You may also describe their Spearman's correlation in order to provide a more in-depth behavioral pattern.
I am not interested in micro-nutrients or caloric intake, but just a
habitual frequency of consumption. The 'snack' category only includes
unhealthy snacks. I was suggesting aggregating by doing a maximum
frequency not simply adding them up. So I wasn't saying that 6 fast
food meals is as bad as having 2 fast food meals + 4 snacks, instead I
was suggesting that eating fast food with frequency 2(once a month)
and eating snacks with frequency 4 (1-2 days a week), by aggregating
them through doing a max of the two would give me a frequency of 4
(highest frequency possible).
Be it maximal or summed, the problem remains there. Your maximal value scheme assumes that having a fast food meal every day is as bad as having an unhealthy snack every day. (Also, you may want to revise your question title, which now says "Adding two categorical variables") Another drawback about picking the maximum is that a case answering 1 & 7 is treated the same as a case answering 7 & 7. This scheme can, at its worst, throw away half of the variability.
At the end, I'd suggest no matter what is done, clearly state the assumptions and defend them in the methods section. Hopefully the people reviewing (if it's a paper to be submitted) the work would agree with you.
