If missing data are not MCAR, then you need to figure out a way to adjust for the non-random impact of missing data on your model.

     My favorite approach is to use the characteristics on which the data are not random to create an estimation for an imputed value.  Basically, you are using the available (full-field) variables to create an equation to predict the variable that has missing data as the outcome.

In epidemiology, we typically refer to this article:

van der Heijden, G. J., Donders, A. R. T., Stijnen, T., & Moons, K. G. (2006). Imputation of missing values is superior to complete case analysis and the missing-indicator method in multivariable diagnostic research: a clinical example. Journal of clinical epidemiology, 59(10), 1102-1109.

For your field, I would suggest searching on the terms "Impute OR Imputation" AND "Missing Values"

  To do the process in SPSS, you use the equation that the prediction of missing data model gives you from your available characteristics to create (Transform, then Compute, then type in the new equation) a new "imputed" variable.  Then you test the values against your existing data variable to see how far off the estimated values are for non-missing fields.  There are also ways to boot-strap the imputed values with repeated models, but that will probably need a program beyond SPSS (or else an advanced module).

If missing data are not MCAR, then you need to figure out a way to adjust for the non-random impact of missing data on your model.

     My favorite approach is to use the characteristics on which the data are not random to create an estimation for an imputed value.  Basically, you are using the available (full-field) variables to create an equation to predict the variable that has missing data as the outcome.

In epidemiology, we typically refer to this article:

van der Heijden, G. J., Donders, A. R. T., Stijnen, T., & Moons, K. G. (2006). Imputation of missing values is superior to complete case analysis and the missing-indicator method in multivariable diagnostic research: a clinical example. Journal of clinical epidemiology, 59(10), 1102-1109.

For your field, I would suggest searching on the terms "Impute OR Imputation" AND "Missing Values"

  To do the process in SPSS, you use the equation that the prediction of missing data model gives you from your available characteristics to create (Transform, then Compute, then type in the new equation) a new "imputed" variable.  Then you test the values against your existing data variable to see how far off the estimated values are for non-missing fields.  There are also ways to boot-strap the imputed values with repeated models, but that will probably need a program beyond SPSS (or else an advanced module).

You can use the MULTIPLE IMPUTATION command to do what Julia describes.  There are many examples online.  See the link below for example--the SPSS example is in the latter part of the document.  HTH.

http://core.ecu.edu/psyc/wuenschk/MV/MultReg/MultReg_Mult-Imputation.pdf

Dear Julia,

Thank you for your answer. The missingness of my data is because that the conduction signal is too weak to collect, The absence of data reflected the severity of the nerve fibers. If the absent data were imputed, then the severity of the nerve would be biased, however, if I don't do this, it can't be used for comparison.

Wish to hear from your comment.

Thanks. 

Danny, the article Julia referred you to is a very nice one.  Here is an excerpt from it that speaks to your concerns (I think):

• Mostly, missing data are neither MCAR nor MNAR [5].  Instead, the probability that an observation is missing commonly depends on information for that subject that is present, i.e., reason for missingness is based on other observed patient characteristics. This type of missing data is confusingly called missing at random (MAR), because missing data can indeed be considered random conditional on these other patient characteristics that determined their missingness and that are available at the time of analysis [9]. For example, suppose we want to evaluate the predictive value of a particular diagnostic test, and the test results are known for all diseased subjects but unknown for a random sample of nondiseased subjects. In this case the missing data would be MAR: conditional on a patient characteristic that is observed (here the presence or absence of the disease) missing data are random.  

If I follow, your missing data is MAR.  And multiple imputation can certainly be used in that case (in the way Julia described, and in the way Karl Wuensch's document shows).  If you have some other concerns, please clarify.  HTH.

p.s. - Alan Acock's article linked below provides another very accessible discussion of the main issues around missing data.

http://onlinelibrary.wiley.com/doi/10.1111/j.1741-3737.2005.00191.x/abstract