It was a small mistake, sometimes that's all it takes.

WARNING: This post will not be funny...  Not one bit.  Unless your idea of funny is schadenfreude.

If you know me well, you know I have a special place in my heart for some really emo music.  Like Bright Eyes level emo.  Deal with it.  (Note: That isn't even close to what this post is about).

One of my favorite B.E. song goes:

So then I fell like a girl from a balance beam.
A gymnasium of eyes were all holding on to me. 
I lifted one foot to cross the other
and I felt myself slipping. 
It was a small mistake. 
Sometimes that is all it takes.


You can hear it here: http://www.youtube.com/watch?v=IHl3ZQbsPDA
If you are so inclined.  

So, I took over this project for a medical student who was no longer interested in working on anything related to research about a year and a half ago.  (Now, I'm wondering whether I should follow in her foot steps.)  I'd been working on it on and off in my spare time.  It is all about shared familial transmission of autism spectrum disorders (ASD) and attention deficit/hyperactivity disorder (ADHD).  According to the results of this project, which is a large, population-based study of ~30,000 families, mothers who have a diagnosis of ADHD are about 6.5 times more likely than mothers without ADHD to have a child with ADHD--not surprising, we already sort of know this. However, mothers with a diagnosis of ADHD are also about 3 times more likely than mothers without ADHD to have a child with an ASD--interesting finding!  Some prior genetic, twin, and family based studies have supported this in the past, but this was the first large population-based study of its kind. 

It should be a really cool study. It isn't really in my area of focus, but still a neat little epidemiological-style study that wasn't too hard to complete. It was submitted to one journal, they gave it a positive review, but they just didn't feel they had room for it.  So, we sent it to another and got a very positive review with a request for some minor revisions, as well as some fun additional analyses we could do to look at siblings.  Now here is the rub...

Siblings?!  I'd forgotten about siblings.  Did we even have sibling data in my data set?!  How would they be labeled?  There certainly weren't siblings separated out by columns, so they would have to be in rows...right?  ROWS?! F*CK!!! This wasn't our original question, so I didn't even really think all that much about this issue.  Someone else just handed me the data and told me "to go to town" after all.

If you do family-based research or nested designs of any kind, you already know where this is going, and I should have known better than to take anything someone said about a data set at face value, but I F*CKED UP!!!

BIG PROBLEM with ORIGINAL ANALYSES—

The info got originally from Med Student showed ~35,073 primary cases for analysis, which is what we used for analyses in the original paper using a selection variable (1=exclude, 3=include). This is what was given to me in the original data set: ~45,000 cases and a selection variable.  Since this was the selection variable I was told to use by the previous person working with this data, that's what I did. WRONG DECISION!
 
I was told this selection variable was created to select children for inclusion in analyses if they were singletons (i.e., only children) or if they were *THE* child selected at random from multiple child families, but it wasn’t.  It wasn't at all!!!

Instead, the selected children in that data set (i.e., those marked as include) were any children that fell in the appropriate age range  for the sample (regardless of whether they were in the same family).

That is...There are only 28,651 families in the data set, so there should only be that many primary cases for analysis…all singletons and all children selected from multiple child families...ONE CHILD PER FAMILY == 28,651 child mother pairs!!!  All mothers only counted one time!

However, in my original analyses--we had 35,073 (all those I was told to include), so we were counting mothers from multiple child families multiple times as long as they had a child in the data set who fell in the correct age range.  This is problematic for issues of independence, as well as our estimates of how many mothers had which diagnoses, because we were counting some of the moms twice, as if they were two different people.

It may also help to explain why some of our prevalence rates were off, etc., but that is WAY beside the point now.

I’ll have to go back and redo ALL the analyses.  ALL of them. ALL.


Not to mention, this means I can't just simply fix the few minor things targeted in the review and add the fun analyses.  It means, I have to redo everything and explain to the editor that I F*CKED up, and now the original version of the paper I sent had to be completely redone, and probably pulled from consideration all together.  If I were this editor, I'd want to tell this Erica-girl to buzz off because she doesn't even know what she's doing well enough to count the number of families correctly.


UGH!!!

I'm so frustrated!  I cannot believe I didn't catch this sooner.  I'm really angry at myself and disappointed, and of course, now riddled with self-doubt.  I'm not usually sloppy, let alone *this* sloppy.  Of course, the immediate place I go in my heart is I don't deserve to be a professor. I clearly can't handle even basic math.
I'm trying to move away from that place right now.  I'm not a machine. I'm an animal. I'm imperfect.


It was a small mistake, sometimes that's all it takes?