Skipped Tickets / Tickets sent back to queue by agents

Based on the error, it sounds like you may be trying to use a formula that only works for the Updates Dataset, on the Support Default dataset.  The different datasets have different attributes available, and the attributes with Changes in the name, only live on the Updates dataset, so that would explain the error. 

Another part of this, is that your formula is not set up to return anything.  While you can have metrics that don't use an IF statement, typically those are doing some kind of mathematical thing to return an output. I think that you will want to use an IF statement for this formula. There's a lot ways to approach this, but this is what I would do, I would make this a custom Attribute: 

IF
[Changes - Field name] = "assignee_id"
AND [Changes - Previous value] > 1
AND [Changes - New value] = 0 
THEN
[Changes - Previous Value]
ELSE
"False" 
ENDIF

I made a lot of small changes so I'm going to walk through this line by line. 
1. The [Changes - Field Name] needs to match the styling of the field names that you see when you apply it as a filter. Some match the actual field name, some, like assignee and group, are limited to weird stuff like "assignee_id" only. 
2. The previous value needs to be a value that the field could have been set to, it cannot be the name of another field. Since we want to capture when agents change a ticket from assigned to them, to unassigned to anyone, and we know all assignee IDs are greater than 1, this will return any time the value is a number greater one, IE actually assigned. This is a bit nicer than "!= NULL" which I've had problems with accuracy and still returning 0 values, depending on the field you're working with. 

3.  [Changes - New value] = 0  is all good, left it alone. 

4. We need to add in what to do if the above statements are all true. If the assignee ID was changed from a number, to 0, at any point in the ticket, then return the assignee_id that was unassigned from the ticket. 

This would let you filter out tickets where the ticket never had any change from assigned to an agent, back to no assignee, and would return the ID of the assignee who was previously assigned. 

You would want to use a metric like that, against the 'D_COUNT(Updates)' metric, to get back how many times tickets were assigned then unassigned, by the assignee ID that was assigned. This would capture if several agents assigned, then unassigned the same ticket repeatedly.