How to Impute missing Values in Survey Data?

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

siva99
Posts: 9
Joined: Thu Apr 09, 2015 4:38 am

How to Impute missing Values in Survey Data?

Postby siva99 » Thu Apr 09, 2015 5:12 am

Hi....I am currently Working on a Survey data......How to find the data is random or Non-random data? (I feel its Non-Random data )......How to Impute Missing values ?......
Consists of More than 250 variables ?I can't delete the data.....

Is there any method to deal with this kind of data....????
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: How to Impute missing Values in Survey Data?

Postby GerineL » Thu Apr 09, 2015 7:38 am

First, just think logically:
Do you just KNOW that data are not missing at random (for instance, you gave certain questions only to males not females)? Then don't impute for those variables?
Does your dataset contain variables that have no observations at all? (you say something about 250 variables but it is not clear to me what you mean). Than you cannot impute data, because there is no information available to base your imputation on.

Second, you can use Little's test to see if your data are MCAR (missing completely at random). If non significant, you can impute data.
https://www.youtube.com/watch?v=22aR9ruSig4
If it is significant, you can go see what variables for instance seem to have way higher missings.

There are several methods to impute missing data, which are dependent on the test.
For instance, for regression you can use multiple imputation:
https://www.youtube.com/watch?v=ytQedMywOjQ
siva99
Posts: 9
Joined: Thu Apr 09, 2015 4:38 am

Re: How to Impute missing Values in Survey Data?

Postby siva99 » Thu Apr 09, 2015 8:26 am

Firstly its an survey data and i cant delete any data.......
Many questions which were are asked contain multiple choice answers..…
For Example- The TV channels I regularly watch (Mulitple options) A NDTV/24 X 7 B Times Now C BBC D Star Sports E AXN
Some people replied A,B,C and some didn’t……I have created separate variable for all these depicting 0 and 1…….

I have missing variables ranging from 5% to 50%......

I have done little MCAR test....Chi square is 592395.029 ,DF is 590454 and sig 0.37
Can you say from is test the data is random or Not-random data......
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: How to Impute missing Values in Survey Data?

Postby GerineL » Thu Apr 09, 2015 9:18 am

siva99 wrote:Firstly its an survey data and i cant delete any data.......
You said that before, I don't understand this comment.
Why would you want to delete data? Why would it be impossible (i.e., what is the causal relation between survey data and the possibility to delete data)? And even if you can't delete data, you can still decide not to analyze something right?
Many questions which were are asked contain multiple choice answers..…
For Example- The TV channels I regularly watch (Mulitple options) A NDTV/24 X 7 B Times Now C BBC D Star Sports E AXN
Some people replied A,B,C and some didn’t……I have created separate variable for all these depicting 0 and 1…….
So are all your data nominal? What type of analysis do you want to perform on these data.
I have missing variables ranging from 5% to 50%......
You say missing variables, but it is unclear wht this means:
- Does the entire variable just not occur in your dataset? What does that mean? is it a mistake? or is it, for instance, because nobody watched AXN, therefore it has the value 0 for all people?
- If you say missing variable, do you actually mean that some people have missings on 5 to 50% of the variables? I.e., it is missing values for cases, not missing variables? If so, do these missings represent actual situations (i.e., if somebody did not watch AXN, they don't get a 1 in this variable) or do they represent missings because you don't know whether or not someone watched the channel?
I have done little MCAR test....Chi square is 592395.029 ,DF is 590454 and sig 0.37
Can you say from is test the data is random or Not-random data......
This would indicate that it is random, however, as I indicated before: don't just fare on statistics, also think logically. I think in your case you have missing data for another reason, but see questions above.

Just to be clear: missing at random would mean that there is no particular reason why some data are missing and other data aren't. If indeed, like I suggested above, Data are missing because the variable is not applicable for a person, this is not random but represents an actual situation.
siva99
Posts: 9
Joined: Thu Apr 09, 2015 4:38 am

Re: How to Impute missing Values in Survey Data?

Postby siva99 » Thu Apr 09, 2015 12:58 pm

Yup most of my data’s are Nominal….My Objective is to fill missing data….
I have to currently Fill/Impute missing values in the variables….I want to know which method is suitable to fill missing values ???

I feel its Non-Random Data as you- same as you mentioned in last line….

I have done Analyze --> Descriptive Statistics --> Frequencies and found how many missing data’s are there in each variable….

////MCAR Test Result/////
Sig = 0.37
EM Algorithm failed to converge in 25 iterations....
siva99
Posts: 9
Joined: Thu Apr 09, 2015 4:38 am

Re: How to Impute missing Values in Survey Data?

Postby siva99 » Thu Apr 09, 2015 1:05 pm

For Example consider 7 variables are there….A survey taken from 4 People

Name Age City Gender Marital Status Religion Newspaper
Arjun 30 Mumbai M M B A,C
Prakash 26 Bangalore M U B
Anjali 32 Delhi F M C
Shahid 20 Bangalore M A A

Religion --- Hindu–A , Muslim-B , Christian-C
NewsPaper ------ Times Of India –A , The Hindu – B , Dna - C


Created Dummy variables :
Name Age City Gender Marital_Sta Religion_A Religion_B Religion_C New_p_A New_p_B New_p_C
Arjun 30 Mumbai 0 0 0 1 0 1 0 1
Prakash 26 Bangalore 0 1 0 0 0 1 0
Anjali 32 Delhi 1 0 0 1 0 0
Shahid 20 Bangalore 0 0 1 0 0 1 0 0

*(I Left space between each variable but its not displaying here)
Gender---> 0-Male 1-Female
Marital Status --> 0- Married 1- Unmarried

*Original Data taken from more 3000 people and contains more than 250 variables……
****I want to know how to fill the missing values in those variable??????
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: How to Impute missing Values in Survey Data?

Postby GerineL » Thu Apr 09, 2015 3:25 pm

It is hard to read your example because it doesn't line up.
you can add an example from excel or spss or whatever, make an image of it or upload the file, that might be easier.


The most important question is: What do you mean by missing? give a concrete example of data that are missing.

Second question is: Why do you want to impute data? What is your goal here? I assume you want to do more analyses afterwards. Is that right? If so, what analyses?
siva99
Posts: 9
Joined: Thu Apr 09, 2015 4:38 am

Re: How to Impute missing Values in Survey Data?

Postby siva99 » Fri Apr 10, 2015 4:47 am

Let me ask my douts........How to find whether the given data is Mcar ,Mar or Nmar in SPSS ?
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: How to Impute missing Values in Survey Data?

Postby GerineL » Fri Apr 10, 2015 7:35 am

little's test, like I said, but it seems like your data is missing because of the way you coded it.
siva99
Posts: 9
Joined: Thu Apr 09, 2015 4:38 am

Re: How to Impute missing Values in Survey Data?

Postby siva99 » Fri Apr 10, 2015 11:07 am

I have done Little MCAR test and Result is written below.....

Little MCAR test : ChiSquare = 593073 ,DF= 590451 , Sig =.013
EM Algorithm failed to converge in 50 iterations

How to interpret the result ?......What is the conclusion ???
Its a Rondom Data or NOT ???
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: How to Impute missing Values in Survey Data?

Postby GerineL » Fri Apr 10, 2015 2:06 pm

As I said before, this statistic indicates that there the missings are random HOWEVER as I said before also, statistics don't trump logic, and I think that your missings are not random but due to the way you coded them.
siva99
Posts: 9
Joined: Thu Apr 09, 2015 4:38 am

Re: How to Impute missing Values in Survey Data?

Postby siva99 » Fri Apr 10, 2015 5:58 pm

ok i got it.....How to process the data ?
Shall i do Hot-Deck Imputation ? Or any other method is thr ???
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: How to Impute missing Values in Survey Data?

Postby GerineL » Mon Apr 13, 2015 7:48 am

in spss, you don't have that much options for imputation, take a look at this video: https://www.youtube.com/watch?v=ytQedMywOjQ

Who is online

Users browsing this forum: No registered users and 1 guest

cron