How is missing data calculated on Likert Scale? I have the following scales:


*

*1 (Strongly Disagree);

*2 (Disagree);

*3 (Not Applicable);

*4 (Agree);

*5 (Strongly Agree);



Questionnaire responses:


*

*Question Item1: 2;

*Question Item2: 3;

*Question Item3: 1;

*Question Item4: 1;

*Question Item5: 4;


If I need to average the above responses into the Excel Sheet, while treating Question Item2: zero (since it is considered missing data), do I need to divide by 5 or 4 items to obtain the average?
$$\text{Average} = \frac{2 + \mathbf 0 + 1 + 1+ 4}{5}\quad\text{or}\quad\text{Average} = \frac{2 + \mathbf{\_} + 1 + 1+ 4}{4}$$
Kindly advise.
 A: Dealing with missing data is a major topic unto itself, with many different available techniques and theoretical approaches. The two options you've mentioned are by no means the only options. However, of the two, the second is definitely better. It's equivalent to simply removing the missing observation before calculating the mean. The first option treats a missing observation as if the subject very strongly disagreed, which doesn't make much sense, in general.
A: While @Kodiologist is correct, your data doesn't look like missing data to me. Not Applicable means the respondent is nether agreeing nor disagreeing the question. It's NOT a missing value because it gives you something about the respondent. In this case, you should divide by 5.
EDIT: It doesn't make sense to use 0 for Not Applicable because it's not a missing value. What about 3?
A: "Not applicable" is not a missing data but a valid information! Imagine that in survey you ask about ages of respondent's children:

Is your son (1) much younger, (2) younger, (3) same age, (4) older, (5) much older then your daughter,
  or either is the question (6) not applicable because you do not have
  either son, daughter, or both?

If you coded "not applicable" as $0$, you'd conclude that people who do not have sons, or do not have daughters have much much younger sons than daughters. If you code it as $7$, then you'd conclude that they have much older sons. If you code it as some middle value, you'd conclude that childless person's children are in the same age...
If you need to calculate average, then exclude those participants who marked "non applicable" from this calculation (you can report count or percentage of such cases) and treat "not applicable" as another dummy variable ignoring the fact that it was a part of the same question.
