how to impute 'not applicable' or 'refuse to respond' options
Locked
how to impute 'not applicable' or 'refuse to respond' options
 
|
I was hoping someone would be kind enough to offer some advice. I have a pain rating variable for a specific physical activity measured with one question ranging from 0 to 10 and then there are two user-missing options: 11 - not applicable (participant
did not engage in that activity - about 40% of the sample at baseline) and 12 - refuse to answer (very few). I need to impute missing data on that variable for time 2, 3, 4 (longitudinal study). I have two ideas:
1. I should first impute 'not applicable' and 'refuse to answer' as two separate categorical no/yes - 0/1 variables and then impute the pain rating as a continuous variable for those whose impute values are zeros for the two categorical variables.
2. Or should I just ignore the two user missing options and impute continuous values for everyone?
Any pointers to any literature sources would also be greatly appreciated.
thanks so much,
bozena zdaniuk
|
Re: how to impute 'not applicable' or 'refuse to respond' options
|
|
Dear Bozena, If you have a dataset in which some questions do not apply to some people, don’t impute them, but carry out separate analyses for the people to
which it does apply and to which it doesn’t. For the people to which the question doesn’t apply, simply don’t include the specific variable in the analysis. Apparently you have two different populations for which different things apply, so in the subsequent
analyses you have to treat them differently with respect to the specific variables in the analysis. Best, Joost From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Zdaniuk, Bozena I was hoping someone would be kind enough to offer some advice. I have a pain rating variable for a specific physical activity measured with one question ranging
from 0 to 10 and then there are two user-missing options: 11 - not applicable (participant did not engage in that activity - about 40% of the sample at baseline) and 12 - refuse to answer (very few). I need to impute missing data on that variable for time
2, 3, 4 (longitudinal study). I have two ideas: 1. I should first impute 'not applicable' and 'refuse to answer' as two separate categorical no/yes - 0/1 variables and then impute the pain rating as a continuous
variable for those whose impute values are zeros for the two categorical variables. 2. Or should I just ignore the two user missing options and impute continuous values for everyone? Any pointers to any literature sources would also be greatly appreciated. thanks so much, bozena zdaniuk ===================== To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
|
Re: how to impute 'not applicable' or 'refuse to respond' options
|
Is it reasonable that a person *would or would not * participate in the
activity in question at the different time repeats? Is it reasonable that a person would refuse to answer at one time point and not at other time points? ----- Art Kendall Social Research Consultants -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Art Kendall
Social Research Consultants |
Re: how to impute 'not applicable' or 'refuse to respond' options
|
In reply to this post by Joost van Ginkel
I absolutely agree. A related problem is when scores are calculated for Agree-Disagree scales (coded 1 to 5) in which some cases have one or more "Don't know" and "Can't answer" responses (coded 9). A case in point is the SPSS file distributed by the UK Data Service for the 2016 British Social Attitudes Survey (BSAS) in which scores are calculated for anyone who answered at least three items (out of 5, 6 or 8) and the scores then rescaled to a range of1.0000 to 5.0000. Because value 9 is not declared as missing in the source items, only (lo thru -1), the target scales all have a significant number of cases with value 9 (labelled as "Missing" but with only values (lo thru -1) declared as missing, some of the scores in the expected range 1.0000 to 5.0000 have also been calculated to include value 9 from the source items. Anyone using the scores as distributed will therefore get misleading results, even if they modify the missing values to (lo thru -1, 9). Because I re-calculated scores only for those respondents who gave an Agree -Disagree answer to all items in their respective batteries, I obtain completely different results. You can see full details in: http://surveyresearch.weebly.com/uploads/2/9/9/8/2998485/british_social_attitudes_2016_spss_file_technical_note.pdf (47pp, 2.14 mb) The Data Manager at Natcen (who run the BSAS) says the derived variables are fine. The documentation gives no details of their calculations and I have tried without success to replicate the exact scores in the file. Discussion continues, but perhaps we shall have to Agree to Disagree. John F Hall [Retired academic survey researcher] IBM-SPSS Academic Author 9900074 Email: [hidden email] Website: http://surveyresearch.weebly.com/ SPSS course: http://surveyresearch.weebly.com/1-survey-analysis-workshop-spss.html Research: http://surveyresearch.weebly.com/3-subjective-social-indicators-quality-of-life.html From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Ginkel, J.R. van Dear Bozena, If you have a dataset in which some questions do not apply to some people, don’t impute them, but carry out separate analyses for the people to which it does apply and to which it doesn’t. For the people to which the question doesn’t apply, simply don’t include the specific variable in the analysis. Apparently you have two different populations for which different things apply, so in the subsequent analyses you have to treat them differently with respect to the specific variables in the analysis. Best, Joost From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Zdaniuk, Bozena I was hoping someone would be kind enough to offer some advice. I have a pain rating variable for a specific physical activity measured with one question ranging from 0 to 10 and then there are two user-missing options: 11 - not applicable (participant did not engage in that activity - about 40% of the sample at baseline) and 12 - refuse to answer (very few). I need to impute missing data on that variable for time 2, 3, 4 (longitudinal study). I have two ideas: 1. I should first impute 'not applicable' and 'refuse to answer' as two separate categorical no/yes - 0/1 variables and then impute the pain rating as a continuous variable for those whose impute values are zeros for the two categorical variables. 2. Or should I just ignore the two user missing options and impute continuous values for everyone? Any pointers to any literature sources would also be greatly appreciated. thanks so much, bozena zdaniuk ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: how to impute 'not applicable' or 'refuse to respond' options
|
|
In reply to this post by Zdaniuk, Bozena-3
I agree with the warnings I have read.
On the other hand, try to avoid analyzing single items when you can analyze latent scores/ factor scores/ composites. In computing those scores from items, you may compute the mean of those items not-missing. Like,
+ COMPUTE NewScore = mean.6(var1 to var9) .
That works best when the items missing are few; and a Missing does not
have an extreme mean where it does exist.
When 40% are missing for an item of note, whatever is done has to be specifically noted in the eventual report.
-- Rich Ulrich From: SPSSX(r) Discussion <[hidden email]> on behalf of Zdaniuk, Bozena <[hidden email]>
Sent: Friday, August 17, 2018 3:12:29 PM To: [hidden email] Subject: how to impute 'not applicable' or 'refuse to respond' options I was hoping someone would be kind enough to offer some advice. I have a pain rating variable for a specific physical activity measured with one question ranging from 0 to 10 and then there are two user-missing options: 11 - not applicable (participant
did not engage in that activity - about 40% of the sample at baseline) and 12 - refuse to answer (very few). I need to impute missing data on that variable for time 2, 3, 4 (longitudinal study). I have two ideas:
1. I should first impute 'not applicable' and 'refuse to answer' as two separate categorical no/yes - 0/1 variables and then impute the pain rating as a continuous variable for those whose impute values are zeros for the two categorical variables.
2. Or should I just ignore the two user missing options and impute continuous values for everyone?
Any pointers to any literature sources would also be greatly appreciated.
thanks so much,
bozena zdaniuk
|
Re: how to impute 'not applicable' or 'refuse to respond' options
|
|
In reply to this post by Art Kendall
yes, people can (and do) switch from not engaging in a behavior to engaging later in the study. In fact, the study intervention is aimed at improving participants' engagement in that particular behavior. The 'refuse to answer' answers change as well, possibly because people get more comfortable answering questions about the behavior after experiencing the intervention.
bozena ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of Art Kendall [[hidden email]] Sent: Saturday, August 18, 2018 7:53 AM To: [hidden email] Subject: Re: how to impute 'not applicable' or 'refuse to respond' options Is it reasonable that a person *would or would not * participate in the activity in question at the different time repeats? Is it reasonable that a person would refuse to answer at one time point and not at other time points? ----- Art Kendall Social Research Consultants -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: how to impute 'not applicable' or 'refuse to respond' options
|
|
I'd like to suggest a different way of looking at the data. At each time point you have three groups of people: (A) those who did not do the activity, (B) those who did the activity and provided a rating, and (C) those who did the activity but did not provide a rating. Your missing data problem is with (C) and you might assume that data are missing at random (MAR). Group B presents a modeling problem. You have a two part model where in people either do or not do the activity and if they do the activity they make a rating.
A key reference is Olsen, M.K. & Schafer, J.L. (2001). A two-part random-effects model for semicontinuous longitudinal data. Journal of the American Statistical Association, 96, 730-745. Older papers (available from the Mplus website: statmodel.com/papers.shtml look at the "Two-part ..." tab) are Petras, H., Nieuwbeerta, P., & Piquero, A.R. (2009). Participation and frequency during criminal careers over the life span. Vazsonyi, A.T. & Keiley, M.K. (2007). Normative developmental trajectories of aggressive behaviors in African American, American Indian, Asian American, Caucasian, and Hispanic children and early adolescents. Journal of Abnormal Child Psychology, 35, 1047-1062. Witkiewitz, K. & Masyn, K. (2007). Drinking trajectories following an initial lapse. In press, Psychology of Addictive Behaviors. Brown, E.C., Catalano, C.B., Fleming, C.B., Haggerty, K.P. & Abbot, R.D. (2005). Adolescent substance use outcomes in the Raising Healthy Children Project: A two-part latent growth curve analysis. Journal of Consulting and Clinical Psychology, 73, 699-710. Spss can't do this type of analysis. I don't about sas or stata. Mplus will do the analysis. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Zdaniuk, Bozena Sent: Saturday, August 18, 2018 4:42 PM To: [hidden email] Subject: Re: how to impute 'not applicable' or 'refuse to respond' options yes, people can (and do) switch from not engaging in a behavior to engaging later in the study. In fact, the study intervention is aimed at improving participants' engagement in that particular behavior. The 'refuse to answer' answers change as well, possibly because people get more comfortable answering questions about the behavior after experiencing the intervention. bozena ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of Art Kendall [[hidden email]] Sent: Saturday, August 18, 2018 7:53 AM To: [hidden email] Subject: Re: how to impute 'not applicable' or 'refuse to respond' options Is it reasonable that a person *would or would not * participate in the activity in question at the different time repeats? Is it reasonable that a person would refuse to answer at one time point and not at other time points? ----- Art Kendall Social Research Consultants -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: how to impute 'not applicable' or 'refuse to respond' options
|
|
thanks so much, Gene! I will look at all those sources.
much appreciated. bozena ________________________________________ From: Maguin, Eugene [[hidden email]] Sent: Tuesday, August 21, 2018 6:42 AM To: Zdaniuk, Bozena; [hidden email] Subject: RE: how to impute 'not applicable' or 'refuse to respond' options I'd like to suggest a different way of looking at the data. At each time point you have three groups of people: (A) those who did not do the activity, (B) those who did the activity and provided a rating, and (C) those who did the activity but did not provide a rating. Your missing data problem is with (C) and you might assume that data are missing at random (MAR). Group B presents a modeling problem. You have a two part model where in people either do or not do the activity and if they do the activity they make a rating. A key reference is Olsen, M.K. & Schafer, J.L. (2001). A two-part random-effects model for semicontinuous longitudinal data. Journal of the American Statistical Association, 96, 730-745. Older papers (available from the Mplus website: statmodel.com/papers.shtml look at the "Two-part ..." tab) are Petras, H., Nieuwbeerta, P., & Piquero, A.R. (2009). Participation and frequency during criminal careers over the life span. Vazsonyi, A.T. & Keiley, M.K. (2007). Normative developmental trajectories of aggressive behaviors in African American, American Indian, Asian American, Caucasian, and Hispanic children and early adolescents. Journal of Abnormal Child Psychology, 35, 1047-1062. Witkiewitz, K. & Masyn, K. (2007). Drinking trajectories following an initial lapse. In press, Psychology of Addictive Behaviors. Brown, E.C., Catalano, C.B., Fleming, C.B., Haggerty, K.P. & Abbot, R.D. (2005). Adolescent substance use outcomes in the Raising Healthy Children Project: A two-part latent growth curve analysis. Journal of Consulting and Clinical Psychology, 73, 699-710. Spss can't do this type of analysis. I don't about sas or stata. Mplus will do the analysis. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Zdaniuk, Bozena Sent: Saturday, August 18, 2018 4:42 PM To: [hidden email] Subject: Re: how to impute 'not applicable' or 'refuse to respond' options yes, people can (and do) switch from not engaging in a behavior to engaging later in the study. In fact, the study intervention is aimed at improving participants' engagement in that particular behavior. The 'refuse to answer' answers change as well, possibly because people get more comfortable answering questions about the behavior after experiencing the intervention. bozena ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of Art Kendall [[hidden email]] Sent: Saturday, August 18, 2018 7:53 AM To: [hidden email] Subject: Re: how to impute 'not applicable' or 'refuse to respond' options Is it reasonable that a person *would or would not * participate in the activity in question at the different time repeats? Is it reasonable that a person would refuse to answer at one time point and not at other time points? ----- Art Kendall Social Research Consultants -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
| Free forum by Nabble | Edit this page |