All data below is faux data, I'm trying to understand the concepts.
In a set of year-to-year simple survey data, what statistical test can I use to assess for significant differences in proportions over time? For instance, if the proportion of women in an organization was 20/400 in 2010, 25/400 in 2012, 23/400 in 2014, 86/400 in 2016, and 100/400 in 2018 - how can I set up and say there was a statistically significant increase in the proportion of women in this organization from 20/400 or 5% in 2012 and 100/400 or 25% in 2018 (assuming there is a significant difference and I can assess for this). Would it be more of linear plot assessment? Do I compare 2012 to 2018 or the overall trend between each time period (the 2016 v 2018 data I imagine would not achieve statistical significance).
What if the responses were pool from several organizations and not uniform from year-to-year. E.g., 50/567 women in 2012 (75% of organizations responded), 75/488 women in 2014 (62% of organizations responded), and 91/425 women in 2016 (50% of organizations responded). Can I assess for statistically significant differences/non-differences over time in this non-uniform data? How?
Any guidance would be appreciated.
This link will tell how to choose a regression type. That decision depends on the properties of the DV.. Time series are very complicated in my opinion, but you have things that you must have for a particular type of model. Here is the regression link to get you started:
. https://www.google.com/search?q=what+kind+of+regression+should+I+use%3F&rlz=1C1CHBF_enUS847US847&oq=what+kind+of+regression+should+I+use%3F&aqs=chrome..69i57j0l2.21374j0j8&sourceid=chrome&ie=UTF-8
Keep reading and you will catch on to all of this. Best wishes, David Booth
You can do the same for question 2. See the R script in attachment.
The test for trends is sign.:
Cochran-Armitage test for trend: data: propWomen2 Z = 5.5894, dim = 3, p-value = 2.279e-08 alternative hypothesis: two.sided
The input is for women (coded 0 now instead of 1):
50 in 2012
75 in 2014
91 in 2016
for men (coded 1 now instead of 0):
517
413
334
years are
1 = 2012
2 = 2014
3 = 2019
Jos Feys , thank you for the incredible effort and providing the specific script, truly above and beyond.
David Eugene Booth, great resource and link, thank you so much
One more comment from me. Plotting your data versus time is likely the best thing you can do and it will help you choose other methods as well. As @Jos Feys indicates there are many methods out there. Best, David Booth
Jos Feys , I have been struggling with something similar over the past week. However mine is not a trend test as such because I have four proportions from the same sample.
I had the initial variable of interest which gives proportions when cut-offs are applied. Then I adjusted the variable (using three different methodologies) and ended up with three other variables which each give their own proportions.
I need to statistically state if these four percentages are different (if possible, which ones) or not. Do you have any suggestions? Feel free to use the same data presented above.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA