That sounds about right. The GPS needs to “see” the satellites in order to work and four satellites is the minimum number for a valid fix, but when it “sees” more satellites it can improve the potential error in the fix by combining all the data. So going outside should improve your fix and going inside degrades it.
By how much it is degraded depends on the GPS itself (sensitivity and number of channels), on the antenna (gain and directionality), and somewhat on the software used to read the GPS if it misses or garbles a reading.
Let’s say you have a nice Garmin or other handheld GPS and it reports that its accuracy is currently 5 meters, which is a typical number and sounds pretty good. That is the CEP or Circular Error Probable and means that 50% of readings are within 5 meters of true location and 50% are not–imagine a circle around the true location with a 5 m radius and half the readings are inside that circle. We don’t know what the CEP is for your case, but let’s look at the error probabilities.
You can continue on with the statistical analysis for this 5m example and 43% of the readings are outside the 5 meter ring but inside a 2X or 10 meter radius ring. Then 7% of the readings will fall between 10 meters and 15 meters (3X). Finally 0.3% will be outside 15 meters (>3x).
So you can always get odd-ball readings with some probability. If you are taking a measurement every second, that is 3600 times an hour and you should expect to see >3X CEP reading around 11 times an hour over a long period.
Most automotive GPS’s also use Kalman filters to track the position. A Kalman filter is way of taking a noisy signal like a GPS position with errors and creating a less noisy estimate of the true value. It does this by first predicting where it thinks you will be at the next measurement based on its internal state. It then compares the noisy measurement to predicted location and decides how to update the internal state for the next go round. The math is complicated but this has the effect of smoothing the noisy measurements.