I would treat NA as missing in analyzing results for the 5-point scale. However, you could compare those with N/A answers to those who gave ratings if you feel that it would be meaningful to compare them. For example, you might ask if men and women are just as likely to find each item NA.

Hi kinfe,

I think  you should omit these items  from your questionnaire.

Regards

Rafique

Thanks all.

You could simply treat them as missing

Dear all, 

I agree with Ulrich, because statistics should interpret reals life and seems useful to consider it as  a category in risk perception  analysis. Then it should be seen as an important information about that issue of risk perception.

Dear Mr. Kinfe

I have faced with similar problem before. I had also NA items which were important for me to be considered in analyzing my data.

Surely, I am not as expert as the people who answered this question above. Just, I am going to share with you what I decided to do ultimately to keep these NA items! 

In SPSS, I used "if..." under the tab of "Transform". In order to compute the items to have the score of that scale, first, I categorized the participants in two groups: 1) the respondents who can answer to all items, defined as "1", 2) the respondents who can not answer to all items, I mean some items are irrelevant for them, defined as "0".

Then, I defined a condition as below:

"if respondent group=1" ---> Mean (all items) ---> named as "e.g. Varible1_1"

"if respondent group=0" ---> Mean (only the relevant items) ---> named as "e.g. Varible1_2"

By doing so for each group you have the mean score of all relevant items.

The result of computing data will be appeared in two columns which can be easily matched in Excel..

But if you are going to analyse your data in AMOS, please be aware that the factor loading values of NA items will be so low in CFA and so, you have to remove them.

Hope it works,

Farimah

Dear Ferimanh,

You say just the opposite of Prof Ulrich think it is correct for the problem of risk perception

I think you are right only in situations where we could analyse that category as a missing...and the usual procedure is = of what you said...but also in those situations we should analyse if the missings are or not random...

Helena 

Dear Prof. Pestana

Thank you very much for your comment. Yes Prof, you are right. But I think they are not real missing data. They are the items which the respondents are not able to answer. If they are considered as the missing value, they will affect the score. Please let me know if  I am wrong.

With regard to what you said (statistics should interpret reals life) please let me to explain what I am think through an example, in regard of measuring marital satisfaction, there is one item which asks the couples whether they are satisfied with the way they each handle their responsibilities as parents, however some of them have no child. So, what is wrong if we compute the mean score without this item for the spouses who have no child. In my opinion, this mean score is a true score of the level of marital satisfaction for this category and in comparable with the other category --who have at least one child. Am I wrong?

I hope I could explain my idea clearly.

Thank you & Best regards,

Hi Farimah,

I think you should funnel your  research question as parents with no child is not your study population. Mean values come from sample and apply results to the target population. how you are going to generalize your results from sample data if comes from different population. I think you should analyze these cases separately if your study design and research question permits you.

Regards

Rafique

Dear Mr. Rafique

Thank you for your reply. Indeed, having a child has not been considered as a criteria of the eligible participants. I will read more about it. 

Thank you & best regards,

Dear Mr. Kinfe

I would like to appreciate you for your useful question that lead me to read more in order to find my probable mistakes.

Best regards,

Thanks all , all suggestions are crucial.

Dear Dr. Rehman

I think it is impossible. Because the respondents who refused or chose NA will get higher score! How it can be justified?!

Hello Kinfe,

The attached link shows a how one study dealt with the issue of a missing questionnaire item. 

Daryn

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1361174/#b13

Dear Dr. Rehman

Thank you for your reply. I agree with Emma. Actually, I am not sure that value "0" is a correct choice for NA items! Indeed, I think they cannot be considered as the missing values!

However, my knowledge is not enough in this regard. So, please let me share my opinion later --when I become assured.

Thank you,

Regards,

It may all depend on what the % of NAs per question are. In fact your NA may be the key indicator, a bit like the Sherlock Holmes "Dog that didn't bark in the night" phenomenon. Given that you're in the realm of "risk perception" , that would indicate to me that there's another risk besides the one you're considering - things like political risk, reputation, or even simply commercial risk. From the moment your NA-sample becomes important enough, I'd suggest you consider that statistically as an entity on its own and see whether you can actually find some meaningful statistical segmentation in that sample - they may be geographical ones (in or out of an influence zone for example), contractual ones (provided you have those data - but it might be an idea to see what type of agriculture we're talking about which might give you a clue) or any other meaningful parameter that you have besides the scales data. Best anyhow.

In my present research/questionnaire.... I also hv the = issue discussed above regarding N/a.  N/a in my questionnaire do not tally with the meaning of missing data or respondent do not wish to circle the score aside n/a because of privacy issue.  It is upon my supervisor guidance that at the end .... that certain part of questionnaires consist not only the score of  0 - 9 but N/a can be chosen by some of the respondent as some of the matters/items cannot be answer because it is not what they are practicing.  The score 0 in my questionnaires refer to "Not practice at all" when the practice is applicable.  My supervisor told me: it is unfair if a person given no choice except to choose 0 - 9 for a item which is not applicable for them.  While score of 1 - 8 shows that a item is gradually increasing it's level of practice and lastly score of 9 = excellent level of practice.   If I use Rasch.... I have no problem to enter the data as  "X" .... however the challenge I am facing now..... I was informed that Rasch cannot solve research question pertinent to hypotesis.

I found this link useful regarding handling of N/A and missing values.

https://www.spss-tutorials.com/spss-missing-values-tutorial/

 Measurement at an item level implies a confrontational interaction between the demand of the amount or what the scale is measuring possessed by an item and the amount of what the scale is measuring possessed by the respondent. Just .like in cognitive measurement, options imply this value for each item.  For the respondent this amount is fixed at the time of measurement and with this the respondent challenges every item in the scale. Each option must evoke different level of this trait/ability/preference, etc. and any option that does not evoke this behaviour  is  useless and should not be made an option. Except is some opinion polls, where it is is necessary to give every respondent  the option of not to being measured by an item there is scientifically no need for options like "not applicable,"  "not sure," etc. In a scale of 1-5 how do we score these options, especially if there is one indicating or imply a "complete absence" of what is being measured? A score 'zero' or a "three" does not make both a scientific or mathematical scene, and as Knife says it biases the results and leads to wrong inference.  

.Actually,  a score ' zero' or 'three' is a huge disparity and cannot be used in Likert scale to measure 'not applicable' (N/A). It is meaningless  under that context.Nenty's response  is scientifically  correct.

The not applicable response already shows that an item is irrelevant. Skip.

I would like to share what my supervisor had shared with me. In some cases, it is obvious..... the "non-applicable" cannot be simply skip. Few months ago when I faced difficulty on this issue, I also got the same suggestion to delete N/A as part of the score beside the 1-5 ordinal scale. However, when I raise my argument : assuming we delete the "N/A" and yet the item post to the respondent is not applicable to their real practice, what is the reliability of the respond if (1) the respondent decided not to circle any score as it is not applicable to them or (2) the respondent decided to respond and choose '0' with the assumption it mean they are not practicing it. By right, how can a respondent just simply decided to circle '0' for not practicing it at all when the item is 'non applicable practicing' to them at all. And some situation, we cannot avoid or identify the sample group to ensure that every item is an applicable item under their present practice. Bearing in mind, item to identify the level of practice is not the same with item that study on a respondent perception/assumption about their belief or satisfaction. Even in the case of study regarding of satisfaction: there are cases..... when the respondent in the process answering the questionnaires.... there might occur that one or few of the items is/are not suppose to be responded by them as the item is coincidence as a situation which they never experience. However it is vital for a study to determine whether or not to include 'N/A' by conducting a preliminary investigation as it we do not wish that some of the respondent will also simply decided to circle 'N/A' when such practice is an item which must be practiced for the selected sample/respondents.

I would like to introduce how some study cannot skip 'non-applicable' item. In my study as earlier mentioned above, I use IRT (Item Response Theory) - Rasch. Although my study is not in the field of education, I will share how in Rasch can resolve issue in the research area pertinent to 'Examination Assessment'. Eg. In certain exam question paper, student are required to answer only 3 questions out of 5 options given. Therefore, a study on exam assessment cannot just simply skip the other 2 options that the students decided not to answer. And certainly it is difficult to find that all students answer the same questions. Furthermore the purpose of the study is to assess the exam paper that had been prepared. So which ever question that had not been chosen by the student/respondent..... that particular question is to be treated as 'Non-applicable'. The analysis use in Rasch is based on the probability concept. Thus, based on this concept.... it can make probability of what would be the response for the 'non-applicable' item in the event the item is being practiced by the respondent. For 'non-applicable' item in Winsteps (one of the suggested research tool under Rasch).. the N/A score is input in the system as 'x' and will taken similar as missing data. In Rasch, this matter can clearly be seen in its output table "Scalogram". Seems I am not the expert of Rasch, I would like to suggest to those who wish to know about Rasch to google and read book on Rasch 'Applying the Rasch Model: Fundamental Measurement in the Human Sciences" by Trevor G.Bond and Christine M.Fox.

https://stats.stackexchange.com/questions/91552/how-can-i-handle-missing-values-when-fitting-irt-model

or type yr search word at https://www.rasch.org/

For those who wish to further extend their analysis to test relationship/hypothesis, it is possible to use the logit person measure obtained from Rasch ..... then imputed into other research tool like PLS or SPSS. Google ..... there some paper published online which integrate Rasch and PLS

Could anyone please help me with this query:

Does N/A include in Likert Scale ? I am using Qualtrics for analysing my data . The likert scale is : very good, good , avarage, poor , very poor, N/A.

Do I have to say in my analysis a five-point Likert scale or do I have to include N/A and say a six-point Likert scale?

Thank you.

Please refer to Multivariate Data Analysis 7 th ed. Hair, Black, Babin, Anderson Chapter 2. pg. 42-64.

Can NA be used as a part of likert scale 0-5 meaning six point and how to handle NA as a ordinal data response?

I think options such as "not applicable" and "no answer" are structurally missing data but depending on our work, they can be meaningful or not. For example, assume there is a question about how satisfied you are with your current car. If we choose "not applicable", it means we don't own a car which can be important information for the research.