The key is that you are to assume the population varies linearly. That means you will be able to model the population with a line using either the slope-intercept form: y = mx + b or point-slope form: y - y1 = m(x - x1)
Using years after 1990, the 1993 population of 4720 is the point (3, 4720) and the 1996 population of 5200 is the point (6, 5200).
First calculate the slope (m) using the slope formula:

where x1 = 3, y1 = 4720, x2 = 6, and y2 = 5200
m = (5200 - 4720)/(6 - 3) = 160
Now, use the point-slope form:
y - y1 = m(x - x1)
where m = 160, x1 = 3, and y1 = 4720:
y - 4720 = 160(x - 3)
y - 4720 = 160x - 480
y = 160x + 4240
In our case, the "x" is the time (or should be "t") and the "y" is the population (or should be P(t)), so:
P(t) = 160t + 4240
Since 2004 is 14 years after 1990, plug in t = 14 to get your estimate of the population in 2004
William W.
I think you meant to plug in t = 14 for 14 years after 1990 instead of 4.10/21/22