How should I deal with missing data, I have remarkable amount of missing data (blank cases) in 3 categories (my total sample size is 275), in my dataset where either:
1- respondents have abandoned the survey half-way (for some variables I have 75 missing value)
2- for some likert-scale type of items "not sure" options/ I already assigned them a value of 0
3- and some skip patterns: some variables of these skip patterns are binary, categorical and some are scale and ordinal.
My question is how I should code them and define them as "Discrete missing values" in SPSS, should I allocate them specific codes and treat them separately or should I leave them blank.
Hello Mina,
Let's start with the easy one.
1. Skip logic is quite common in surveys, and represents ignorable missingness. That is, it wouldn't make sense to ask a person how much s/he spends annually on vehicle maintenance if they don't own or drive a vehicle. So for such cases, choose a blank or discrete coding value that you will subsequently declare as "missing."
2. For Likert-type response items, avoid choosing a coded value that represents a valid response if no response is given. Either leave the variable blank or assign a discrete, and non-overlapping value to denote that there is no response given. If that is zero (0), fine. If it's some other value (e.g., -9), fine. Just be consistent and do not presume that a missing response should be set equal to the scale midpoint.
3. Respondents quit part-way through the survey. This can be exasperating, but remember that you are completely dependent on the good graces of participants to take their valuable time to furnish data for your research aims. So, when not all elect to do so, don't take it as a personal affront. In this instance, the missing responses are not ignorable.
What can you do in such cases? (a) if possible, follow up with partial completers to see whether they'd be amenable to furnishing the remainder of the responses; (b) Use the partial responses for addressing questions about the variables to which the persons have responded; (c) use listwise deletion, which means retaining only cases having complete response sets; (d) check to see whether these folks are systematically different from those who do complete the response set on any of the variables for which you have information for both. If not, then you have a variety of options available for imputation of missing values. If so, you'd be better served by modeling the respondent group (completers vs. non-completers) as an explanatory variable for whatever analysis you undertake. Each option has some virtues as well as some drawbacks.
Good luck with your work.
Psychologically the missing ones might be due to the fact that they have no idea of the question's message or they are unsure. Balance should therefore be observed in the course of addressing the issue. If it is just an item having responded to other items then on a 5-Likert scale the item should be coded Neutral or Undecided which is 3.if it is a rating scale then often which is coded as 3 should be selected.
You at least three coices: 1- analize data as they are, 2- use equation to estimate missing values, 3' use the averages of values have and fill in blanks! All case are good. I used to use third one. Cheers.
Dear David Morse , Medhat Elsahookie, Busari Yusuf
Thank you very much for your advice, I appreciate your time and help, another note that might help, my total sample size is 275 and for some variables I will have 75 missing values if I include the incomplete questionnaires. Considering this issue what's your suggestion?
- Badges
- Science topic