measuring error rates

#1
Hello - I've got a very basic question, but hoping for some help. Not even sure if this statistics or just algebra. I am trying to determine the error rate for a parking occupancy survey. Surveyors walked many blocks at different hours of the day and counted cars parked at legally available spaces (this availability changes throughout the day). The first col. is the time of the run (8am through 9pm), the second col. is the actual number of legally available spaces, 3rd and 4th cols. are what they survey counted as legally available. I need to calculate the error for the time period, for the block on the specific day, for the entire route for the specific day, and for both days combined.

Example:
Time Actual 14-May 16-May
block 105
8 14 12 17
9 14 13 17
12 13 14 17
13 16 16 16
16 17 17 18
17 16 17 17
20 15 16 18
21 15 17 18

block 104
8 12 11 12
9 13 14 12
12 13 13 12
13 13 13 12
16 13 12 12
17 13 12 12
20 13 12 12
21 13 13 12

block 103
8 12 12 11
9 12 12 12
12 12 12 12
13 12 13 12
16 12 12 12
17 12 12 13
20 10 12 11
21 10 11 12
etc...

I've simply counted the number of times their count does not match the actual (i.e. (8 run - 6mistakes)/8runs) to determine how often mistakes were made. This lets me know how often mistakes were made, but not the degree of the error. I've looked at total inventory for the day for the block, and used the same calculation I used for error frequency. This gets closer to capturing the degree of the error, but if someone made an error of +4 on one run and -4 on another it appears as though no errors were made in the total. I've looked at standard error to get at the dispersion.

I'm not sure what makes the most sense here.

I realize this is basic, but any help would be appreciated!

Thanks!
 

HiLo

New Member
#2
Hi,
One approach I have found useful in similar applications is Mean Squared Error (MSE). For the MSE, calculate the error (+ or -), then square it, for each observation. Then, take the average over the dimension of interest (the block, time period, etc.). The MSE is advantageous in that it gives greater weight to larger differences. I don't believe it has any interpretive value in itself, but it might be a good way to make relative comparisons across the dimensions of interest.

HTH
HiLo