Hello all,
I am currently conducting an aviation capstone project and it has a been a while since I analyzed data/statistics. I am attempting to determine (if/what) statistical analysis is appropriate. My first research question asks:
Are Part 91 (for compensation or hire), 135 (charter) and 121 (cargo) operators more prone to fatigue-related accidents than Part 121 passenger carriers?
With this question the data I have accumulated is in the form of summary statistics. Number of accidents that are fatigue related and total number of accidents. Each of the four categories(Part 91, 135, 121cargo, 121passenger) has a 'fatigue related and total accidents' column.
DV1: Number of fatigue-related accidents
IV1: Part 91 commercial operations
IV2: Part 121 (non-passenger) commercial operations
IV3: Part 135 commercial operations
IV4: Part 121 passenger commercial operations (protected by 14 CFR Part 117)
Would this lend itself to itself to any sort of statistical analysis? Or is this better off remaining as descripitive vs inferential?
My 2nd RQ is :
Is there a particular phase of flight where fatigue-related accidents occur more than others?
For this one I have multiple IV's and DV
DV1: Number of fatigue-related accidents
IV1: Taxi
IV2: Takeoff
IV3: Climb
IV4: Cruise
IV5: Approach
IV6: Landing
I chose MANOVA for the second RQ based on the data I have available for that is much more in-depth. I attached picture of my data in table form.
I would greatly appreciate some guidance on which direction to go! I am a little lost
Thank you so much in advance!
v/r,
Eric
Hello Eric,
Since your data are frequencies/counts, (m)anova isn't really a good analysis method.
For RQ1, a chi-square test of independence (4 x 2: 4 rows of operator category; two rows for fatigue-related count, non-fatigue related count) would address whether operator category relates to likelihood of fatigue-related accident (among all accidents). Note that non-fatigue related count would be total accidents for an operator category minus fatigue-related accidents for the same category.
For your data, the answer appears to be yes: Chi-square(3, N = 16,050) = 164.489, p < .001; there is a relationship. The strength of association (using the Phi coefficient), is .10 (on a scale from 0-1.0).
For RQ2, using the pooled data for the 7 phases (6 if you omit "other" as a flight phase category), I'd suggest using a chi-square goodness of fit test to determine whether fatigue-related accidents appear equally often across the phases.
For all 7 phases, there is a difference, chi-square(6, N = 358) = 102.788, p < .001.
For the 6 phases (with "Other" excluded), there is a difference, chi-square(5, N = 262) = 66.626, p < .001.
Descriptively, categories showing larger misfits to presumed equality included: Landing (higher than overall average), Taxi and Climb (lower than overall average); Takeoff and Cruise/Maneuver (somewhat higher).
When "Other" is included, it showed the largest misfit: much higher than overall average of incidents.
Good luck with your work.
Hello Eric,
Since your data are frequencies/counts, (m)anova isn't really a good analysis method.
For RQ1, a chi-square test of independence (4 x 2: 4 rows of operator category; two rows for fatigue-related count, non-fatigue related count) would address whether operator category relates to likelihood of fatigue-related accident (among all accidents). Note that non-fatigue related count would be total accidents for an operator category minus fatigue-related accidents for the same category.
For your data, the answer appears to be yes: Chi-square(3, N = 16,050) = 164.489, p < .001; there is a relationship. The strength of association (using the Phi coefficient), is .10 (on a scale from 0-1.0).
For RQ2, using the pooled data for the 7 phases (6 if you omit "other" as a flight phase category), I'd suggest using a chi-square goodness of fit test to determine whether fatigue-related accidents appear equally often across the phases.
For all 7 phases, there is a difference, chi-square(6, N = 358) = 102.788, p < .001.
For the 6 phases (with "Other" excluded), there is a difference, chi-square(5, N = 262) = 66.626, p < .001.
Descriptively, categories showing larger misfits to presumed equality included: Landing (higher than overall average), Taxi and Climb (lower than overall average); Takeoff and Cruise/Maneuver (somewhat higher).
When "Other" is included, it showed the largest misfit: much higher than overall average of incidents.
Good luck with your work.
David Morse Jos Feys ,
Thank you so much! This has got me on the right path now after spending some time going over the method. I really appreciate the help with this!!
v/r,
Eric
Just for fun: I added question 2 to the R script. The results, of course, are the same as those from David. I included Post Hoc multiple comparison tests (with Benjamini & Hochberg corrections) as described in the "R companion": https://rcompanion.org/rcompanion/b_05.html
One can run R scripts online at: https://rdrr.io/snippets/
Folks,
Thanks again for the help thus far! I have one quick followup question regarding 'expected' frequency for my second dataset. My null hypothsis states that there is no difference in the phase of flight for fatigue related accidents, or that they are equally distributed. Is it safe to utilize 1/6 or 1/7 of my N=262 to determine an expected freq.? Removing 'other' I'm left with ~43.66. There should be no reason why I couldn't have the same expected frequency for every category in this case, correct?
Thanks again!
v/r,
Eric
- Badges
- Science topic