variance in measurements

I am using a NI USRP 2921 to analyze utilization of WiFi for research. The program I am running brings in dB values and compares them to a threshold to 
make sure it's not noise. I then interpret the data to get a duty cycle. I've ran several test and got varying results from 0.5% error to 25% error changing 
nothing in the program. there is a large amount of variance, Any assistance would be appreciated. If I need to give more information I can do that.