There are many things to note in these data. First, the median of the data were significantly lower than the mean, so it is a very bottom-heavy data set, making the gradient hard to see. The reason for this is not clear, but I put forth a suggestion as follows:
Data from the Bureau of Labor Statistics indicate that the average tenure is around 4.5 years during the time under consideration here. The data from LendingClub suggest that the average tenure is only 1.5 years. This dramatic discrepency might have a few causes.
The LC tenure data are self-reported, and have financial consequences, but this might result in the client giving a larger number, not a shorter one.
The BLS data are for all age groups, but the younger age groups do align with the LC data. Perhaps the demographic for LC skews younger. The LC data do not include ages, so this can't be confirmed.
Some states have markedly fewer data points that could result in a deviation from other states' averages. This was checked, because it could be, and some states definitely suffered from this problem. For example, North Dakota only had 38 data points, Maine had 111, Iowa had 156, Tennessee had 481 and Mississippi had 577. Most other states had 10,000 or more points. This is certainly the cause of bimodality.