it depends of the exension of the missing. if you have more than 10% of missing in your measures, exclusion it is a possibility. If you don´t, try to replace it. However, first you should certificate that missing values are MCAR (missings completely at random)
More info is needed in order to advise, and particularly why they are missing. If it is a big sample with only a few missing, then what you do should not have much effect. If the missing are due to say, the photocopier not printing on one side, that is different from say people without children leaving questions about their kids school blank, and both of these are different from say people not reporting deviant behavior on a questionnaire handed out by their teacher. In general, imputing is design for MAR, but this is a tough assumption to test.
Why would you delete cases with missing values? They are likely to be different to cases with complete values, so you are building bias into your sample by so doing.
Investigate the missing values! We had a stigma questionnaire that had a missing value problem. By examining patterns of missing values we quickly realised that some of the questions referred to the experience of stigma in the workplace or in job security. And of course some of the participants had never worked or were long-term unemployed.
We re-scored the measure omitting the items that didn't apply to everyone.
So before you impute missing values, ask yourself why they should be missing.
But never throw anything away! (Sorry – my grandmother was Scots…)
If you are using SPSS it will use listwise deletion as the default method for missing data which basically excludes cases that have missing data from your analysis. However, this method produces bias if the data are not missing completely at random (MCAR). You can find out if your data are MCAR by using the Little's MCAR test. I am not familiar with the expectation maximization method as I tend to use Multiple imputation which is suggested to produce unbiased estimates when the model is correctly specified.
I would suggest that if you have a large enough sample size and your missing data are MCAR to go with listwise deletion. Otherwise it is better to replace the missing data