Hi,
Suppose, "intention to purchase product x" is a dependent variable (DV) in a particular study. There are several other independent variables (IV). The objective is to examine the effect of IVs on the DV. A respondent selects negative response (e.g. disagree/strongly disagree) against all the items of that particular DV. Now, there are two options to be considered during purification of data.
1) discard data of that respondent since he/she is not a potential customer.
2) keep the data for analysis since unwillingness to purchase might be significantly affected by the effect of various IVs in the study.
Any suggestion?
Hello Sujit,
If your sample size is large, compute the standard deviation for all the answers of each participant (while transforming an ordinal variable (Strongly Agree - Agree - ...- Strongly disagree) to interval variable (1 to 5)).
If the standard deviation is lower than some threshold you pick, remove the participant from the experience, as they seem uninterested in the questionnaire and replying with the same answer always.
Best of Luck,
Halim
What if the participant truly answered no to all of your of questions.
There is no ultimate answer for this question, it will finally be up to the statistician to decide.
Reframing your example, how will you deal with the respondent who checks "Agree" and "Strongly Agree" on all the IV questions? Your option #1 suggests you should delete this respondent too because she/he is definitely a customer!
Logically, respondents who see brand features (for example) as bad will express low likelihood of purchase, and respondents who see those features as good will express high likelihood of purchase. I would think you would want to include such respondents.
Halim Tannous makes a very good point. You should not be concerned so much with whether someone rated IV's mostly high or low, but with the range of scores used. Most inferential statistics are used to see how much variation in one or more variables affect the variation in another variable. If there is no, or very little, variation in your data, then you won't get useful results. So you would want to ensure that your respondents used a reasonable range of values in their responses. (This is another reason why Likert scales items, when used as single IV's, have limited value - but that is another story/rant.)
Hume F. Winzar " Reframing your example, how will you deal with the respondent who checks "Agree" and "Strongly Agree" on all the IV questions? Your option #1 suggests you should delete this respondent too because she/he is definitely a customer! "........this portion is not clear, how is this related? if you would explain. The problem is related to the responses associated with DVs only. Whether an individual is a prospective customer of a particular product or brand, can be defined based on the responses associated with IVs?
I'm sorry, Sujit Kumar Ray . I misread the question. I should have said, " how will you deal with the respondent who checks "Agree" and "Strongly Agree" on all the DV questions? Your option #1 suggests you should delete this respondent too because she/he is definitely a customer! "
The crux of the qestion remains the same: If you are considering removing respondents who answer at one end of a scale, shouldn't you also remove those who answer at the other end?
Hume F. Winzar The problem is not related to extreme responses. If a respondent checks "agree" or "strongly agree", he/she appears to be a potential customer. On the other hand, if he/she checks "disagree" or "strongly disagree" on all the items, he/she doesn't appears to be a potential customer. The objective is to check influence of particular IVs on the DV. Here the DV is potential consumers' purchase behavior which is represented by the construct of "intention to purchase". The confusion is: if I want to examine influence of IVs, should I remove respondents with "only negative" responses to all the DV items? or I should consider them since unwillingness may also be outcome of effect of IVs.
If you want a proper measure of the relationship between your IV's and your DV (Purchase Intention) then you want a DV with as much systematic variation as you can get. That means including respondents who have low purchase intention as well as respondents who have high purchase intention.
Removing respondents who score low is not helpful, or it's just as helpful as removing respondents who score high.
Now, it may be that you are trying to answer two different questions...
If that is the case, then you should have two separate measures.
- Badges
- Science topic