It is difficult to understand this otherwise simple idea because of the weird bias technique used in the input differential stages of op-amps. While the classic biasing is *by the side of the input (base)*, here the biasing is implemented *by the side of the output (emitters)*. This is another topic but still to mention that this trick is possible here since the voltage of the common node between the joined emitters is fixed at the differential mode (we cannot use it in the single common-emitter stage since the emitter voltage will follow the base voltage and there will be no amplification).

So, the biasing constant current source makes the transistors adjust their common emitter voltage so that to pass the 1/2 bias emitter current. For this purpose, they adjust their *beta* times smaller base currents that are produced by the same emitter current source.

But these *input bias currents* have to go somewhere. And designers have chosen an unusual solution - *to pass currents through the input voltage sources*. For this purpose, they have to be "galvanic"; if they are not, they must be shunted with high-resistance elements to ensure a path for the bias current.

So, this is the situation - *input bias currents flow through the input voltage sources and their internal resistances*. If there are additional resistors in series (as in the case), bias currents will flow through them as well.

As a result, bias currents "create", according to Ohm's law V = I.R,  voltage drops across resistors. They are constant since both current and resistance are constant. So, we can think of this resistors as of "batteries" with constant voltage that are connected in series to the varying input voltages. Depending on the polarity, these voltages will be added or subtracted to/from the input voltages; thus they "shift" the varying input voltages with some small *constant* value. And we can adjust the voltage "produced" by these "batteries" by changing the resistance (we cannot change the current since it is set by the internal bias current source in the emitters). 

In the OP's circuit, a voltage drop is created by I- across R1 that is added to Vin-. To compensate it, we have to add the same voltage drop to Vin+; so we include a resistor R3 with the according value in series to Vin+.

So, this is a simple electric arrangement of two (voltage and current) sources and a resistor where ***the combination of the current source and resistor can be thought of as another but constant voltage source in series to the varying input voltage source***.

This circuit solution is used in internal op-amp structures (e.g., in Widlar's 709) to "shift" the voltage variations at the output of the input stages.