Cleaning up gate count statistics: summary

[Laura Baker]

I want to thank everyone who replied to my February 14 post "Cleaning Up Gate Count Statistics" in which I lamented the poor data we had from our gates, the effort it takes to clean it up, and the overall inaccuracy even after clean up.  I wanted to summarize the replies for the list.

Many people commented that poor data is a problem, especially with gates where you physically have to record the meter into a spreadsheet.  I was relieved to hear that we were not alone in this experience.

If just a day or so's data is bad or missing, some people recreate it by interpolation:

"I looked at proportional trends for the affected days that you know had good data. After collecting these data for 15 years, I’ve found that there are almost predictable proportional trends to these data. To elaborate… Look at similar or adjacent weeks with good data, add a similar period together and calculate the proportion (%) that applied to those similar periods to the product of your bad day. For example, if a Wednesday had 10,000 entrances and Thursday had 11,000 entrances (reliable data), then calculate the proportion for each day (in this case W=48% and Th=52%). You could then apply these proportions to the total that was calculated for the days that had transcription errors. So, if the product for the W-Th with questionable data had a total of 22,038 then I would interpolate the W to be (22,038 x 0.48) 10,494 and Th would be (22,038 x0.52) 11,544."

Many recommended developing data validation formulas on the spreadsheet to catch basic errors at the point of entry:

"If the gate count on the 2nd is greater than or equal to the gate count on the 1st , report back gate count on the 2nd, else return the average of the 1st and the 3rd. =IF(A2>=A1,A2,(AVERAGE(A1,A3)))"

Several recommended new gates that offer more accurate ways of counting people and that automatically send data to a server so it does not have to be entered manually.  Many people said new gates solved much of their data problems, although they were not without unique problems of their own.  Gate systems mentioned:

Density

raspberry pi options (https://journal.code4lib.org/articles/12947)

Trafsys

Sensource

Finally, a couple of people reflected on whether gate counts are the best statistic at telling the library story in the first place.  It was a reminder that sometimes we have to consider how much effort to spend trying to tweak flawed data in order to focus our time on more meaningful measures. 

I will certainly talk more with our library administration about how we use gate counts stats, mentioning some of the newer systems if we want more accurate data.  In the meantime, I'm working on a more streamlined form to collect the data that will use some data validation to cut down on the errors.

Thank you, everyone, for your help.

Laura

--

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Laura Baker
Librarian -- User Experience and Assessment
Abilene Christian University Library
221 Brown Library / ACU Box 29208
Abilene, TX  79699-9208

bak...@acu.edu
phone: (325) 674-2477
fax:   (325) 674-2202
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~