Dear
We apply a survey (with 1000 informants and 31 questions) using a likert scale (of 7 points).
As expected, we have several missing values. We'd like to know which methods you believe are the most appropriate to impute values to these missing values. We thank everyone for their answers.
Hi Georges Flexor ,
You can handle this situation by applying following techniques.
1. Neutral-value substitution: You can use neutral value of the likert scale. In your case neutral value would be 3.5 but if your overall average score is less than 3.5 then you could not able to use this method.
2. Mean-value substitution: If the average response for your actual respondents is 2.67, then substitute 2.67 for all your missing values. If it’s 5, use 5. But be aware that it is applicable only when the number of respondents with missing data and the number of items missing were 20% or less. More here Article Missing Data in Likert Ratings: A Comparison of Replacement Methods
3. Imputation: You can calculate those values by the help of SAS and SPSS by using Approximate Bayesian bootstrap with Propensity score. More here Article On the Imputation of Missing Data in Surveys with Likert-Type Scales
Hope this will help you. Good luck.
SAFE, SOUND AND SCIENTIFICALLY ACCEPTABLE METHOD is to remove the defective survey. A survey that s incomplete is considered defective. Remove it. Calculate you minimum sample size. Make sure that after removal defective survey, you can still meet minimum sample size requirement. Do a pilot test to see the % of defect so that you can gauge how many survey to collect and would still meet minimum sample size after removing defective (incomplete) surveys.
SURVEY DATA by definition is the data that has been "collected" by survey. The data was not created by substitution or imputed. The survey data should come from the responses given by respondents. Any survey containing incomplete data is considered defective. the researcher must take the time and effort to supervise survey collection, i.e. make sure that the survey is completely filled out by each respondent.
SOLUTION to a problem is the removal of the cause of the problem, not compounding the problem with ethically questionable methods of "making up" data.
By deleting "defective survey", I believe that Paul Louangrath means not including any observation that includes missing data. This is a very conservative approach, and instead I recommend the options suggested by Dhirendra Nath.
The only way would be to get rid of such surveys where the respondents havent responded.Putting in any value will be a manipulation of data
Thank you for your response. It's just hard to throw away an entire questionnaire when just one or two pieces of information are missing!
I followed the recommendations of David L Morgan & Dhirendra Nath
Thanks to all of you
- Badges
- Science topic
- Similar topics
- Statistics
- Survey