Comparison of 12 measurements in 2019 and 12 measurements in 2020 of the number of individuals in a facility. What is the best method? I need to compare attendance measurements within the same facility, taken each last day of the month in the year 2019 and the year 2020. So I have 12 measurements in 2019 e 12 measurements in 2020.
The question I am trying to answer is whether there was a significant reduction in the average number of individuals present within the facility in 2020.
Is it a paired t.test sufficient to answer this question or a non parametric test like Wilcoxon Mann Whitney would be better? Or am I completely off the track?
Thanks in advance. As maybe you can tell I am new to the field of statistics, so you will forgive a basic question.
These are the data




month
2019
2020




Jan
778
871


Feb
787
891


Mar
818
781


Apr
803
690


May
827
656


Jun
855
674


Jul
853
706


Aug
842
822


Sep
855
735


Oct
879
722


Nov
870
731


Dec
851
671



 A: There may be a seasonal effect, so you should consider a paired test.
If the 12 paired differences d = x.19 - d.20 are nearly normal,
then a paired t test should be OK. But of there is a Covid effect after the
first few months of 2020, data may not be normal. In that case,
perhaps data will be far from normal, and you might want to use
a nonparametric Wilcoxon signed rank test on the differences.
If the point is to illustrate a Covid effect maybe compare only
nine months April through December for the two years.
If you know that there were attendance restrictions because of
Covid, so that attendance was greatly reduced, you might not need
any statistical test to confirm the obvious.
If you want a less-speculative answer, perhaps post the data.
Fake data to illustrate:
x.19 = c(521,526,552, 560,548,436, 388,290,383, 490,524,517)
x.20 = c(532,529,650, 334,357,240, 211,205,241, 220,346,309)
d = x.19 - x.20

summary(d)
  Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 -98.0    63.0   177.5   130.1   199.0   270.0 

A boxplot shows marked skewness and a median very far from 0.
boxplot(d, horizontal=T)


A normal probability plot is far from linear. With such a small
sample size I would not trust to the P-value of a paired t test
to be accurate.
qqnorm(d);  qqline(d, col="green")


t.test(d)

        One Sample t-test

data:  d
t = 4.0056, df = 11, p-value = 0.002066
alternative hypothesis: true mean is not equal to 0
95 percent confidence interval:
   58.60574 201.56093
sample estimates:
mean of x 
 130.0833 



wilcox.test(d)

        Wilcoxon signed rank test

data:  d
V = 71, p-value = 0.009277
alternative hypothesis: true location is not equal to 0

There is a clear difference between the two years, which may not
require a test to be obvious. For what it's worth, I would prefer the nonparametric test.
