Coefficient of Logistic Regression Fitted a logistic regression to predict if a member would default the loan payment. The coefficient for 'loan_amnt' is positive as shown below:

As it is positive, an increase in loan_amnt would mean that odds/probability of 'defaulting' also increases.
But, the average defaults based on quartile of loan_amnt indicates that average number of defaulters decrease with increasing loan_amnt.

columns(quartile, no of members, no of defaulters, average). Here the members is classified as Defaulter or Non-Defaulter.
How can they different ? 
 A: Thought experiment.  Suppose you take five identical copies of one person, and lend each of them 10, 100, 1000, 10000, and 100000 dollars.  Which copy would be more likely to default?
Now suppose you go to a bank and find (different) people that were loaned 10, 100, 1000, 10000, and 100000 dollars, who would be more likely to default?
In the first case, the more you loan to the person, the more likely they are to default.  Since you have 5 copies of the same person, their means to repay the loan are fixed, so the more you loan, the more stress it puts on their fixed finances to repay.  This is what your regression coefficient is capturing.  It is the rate of change of (the log odds of) default as the load amount is varied as all other measurements are held equal..
In the second case, the people you find that the bank has loaned more to will have better financial means than those receiving small loans.  This makes them more able to weather financial stresses, and less likely to attempt to spend out of their means.  This is what your summary table is capturing.
A: Confounding variables!
I don't have the time to sketch out a longer answer, but the basic idea is that you have confounding variables.
Just to make up an illustrative example - Imagine that having a higher loan amount (all else being equal) increases your default rate, and that being very wealthy (all else being equal) decreases your default rate dramatically.  
The regression model tells you that, "on its own" (in a rough sense), a higher loan amount increases the default rate.  That's how you get your first result
However, those who have a higher loan amount might also tend to be richer.  Each of your quantiles might be so much richer than the last, that the effect of being richer (driving down default rate) offsets the effect of having higher loan amounts.  Something like this could explain your second result.  
