are you a math wizz, correspondence analysis

[jessica rabbit]

i need a little help from all you super intelligent men and women out there.

i am trying to do my dissertation for uni, and needed to do a correspondence analysis on my frequency data. i am having a little trouble interpreting my graphs though. i dont know which points are classed as an association to the impact direction of a vehicle and which are not, how do i work ths out?

any help would be more than greatly appreciated.

please please please help.

and if any of you are tutors at my uni, please ignore this post its hallucinations from the diesel fumes.

[POPEYE]

I'm afraid i aint got the faintest idea what your talking about, but there again i'm secondary modern and borstal poly tech

[gwh200]

not to sure about what you are trying to achieve, as all impacts semm to be related to an associative result, but some questions first

1. what do the axis correrspond to,they have no labels and i cannot think of a relation you would deduce from them.

2. more of a moot point,but is this the right graph to present this cumalative data?
i would have thought that the only way (assuming you are using excel) you can present data in two modes,ie impact( assume x) and its relationship with injury (assume y) is to insert a best fits line,but as your data is all over the graph im hazarding that the graph i s the wrong choice.
as for analysing the data into injurious/noin injurious,not quite sure what you are trying to achieve,please simplify your argument cos im feeling thick

right i think i know what you mean.

its a simplified version of that (x-xaverage)/(y-yaverage)

thingy.you dont need to find a least fits line but you do need to take the second table you have and sum and average each column.
once you have done this you ned to profile each point with relation to its relative frequency(the amount of occurences for a specific accident i guess)

but if you just sum and total the columns that will give you the total,cumalative and average results.you can go further and sum and square the results and apply the (x-xav/y-yav) thingy to get definitive answers with out having to bother with things like quartile ranges for extreme results.
give us ten minutes to look at it and ill post atable of results with what you are (hopefully)looking for

[Apache]

I cant read your graph, the resolution is too low.
I'd hazard a guess that the Y axis is related to impact direction, and possibly skull injuries (for instance) happen in head on impacts? X axis is a mystery, unless +/- indicate a deviation from an average figure?

<edit> Agree with Gra. Need more info.

[yoshie]

The graph is quartered in each you have an impact direction one quarter has two. Within these you have the indicated injuries sustained.
This is how to read impact realated to direction on that graph. However, I agree with Graham a better graph choice would help

[jessica rabbit]

hi thanks for replying

the x axis is dimensionon 1, the y axis dimension 2. the data, according to the stats tutor is catagorical. the stats tutor said this was the best way to show the associations in this data. the correspondence analysis graph is a multidimensional graph that shows the associations in a 2 dimensional plot.
i am trying to see how good the associations are between the injuries sustained in car accidents to the occupant seat position, using the impact direction.

its suposed to be that the injuries that are located close to the impact direction plots are related, i just need to know a way to tell how well they are related.

[gwh200]

as of edit above,the answers dont lie within the graph but as a cumalative set of results from the table,will post up some basic results in a mo

[jessica rabbit]

as of edit above,the answers dont lie within the graph but as a cumalative set of results from the table,will post up some basic results in a mo

ta very much

[LURCH]

Have just logged on to the wrong site or sumit, ain't got a bl00dy clue what your on about, but i like your avatar...............

[jessica rabbit]

it might be worth pointing out that there is more than one of these graphs. i am trying to see if from injury types i can place a person in a particular seat of a vehicle.

i have been told to use the correspondence analysis test, as i am checking more than one seat position for associations. the one posted is just for the driver.

i know that the skeletal injury plots that surround the impact direction are related, i just need to know if there is a way of checking how related they are. and then use these CA graphs to show if there is a way of identifying the seat position from injury data.

i need to keep this really simple, as my dissertation is due in little over a month, and its half written already. i just need to know how to read the graph, if anyone has any ideas. i know this type of graph is above the level an undergrad needs but it was recommended to me for use with this data.

i am also not a maths wizz by any stretch of the imagination lol

[jessica rabbit]

Have just logged on to the wrong site or sumit, ain't got a bl00dy clue what your on about, but i like your avatar...............

ta very much, its a factual representation of me lol

[gwh200]

there isnt to much to glean from this as there are only one set of data points per "accident". it may be beneficial to correlate an injury not just with impact direction but also to show less injuries not in an impact direction,but you will have to produce another table of results to show in effect what "didnt" happen.you can use this table though.
info i can glean from this tells me ,obviious data ie 151 frontal injuries which account for 18.875 percent of injuries,which when viewed relatively to the graph will show what % of injuries occur in the front of someone involved in a frontal acident.its all there you will just have to do a bit of counting

good luck lady !!!!

edit, cant add excel table for some reason,but sum the columns then average the columns then this will give you correspondence

[jessica rabbit]

this is the maths that the R program gave me. hope ya can see it

[jessica rabbit]

there isnt to much to glean from this as there are only one set of data points per "accident". it may be beneficial to correlate an injury not just with impact direction but also to show less injuries not in an impact direction,but you will have to produce another table of results to show in effect what "didnt" happen.you can use this table though.
info i can glean from this tells me ,obviious data ie 151 frontal injuries which account for 18.875 percent of injuries,which when viewed relatively to the graph will show what % of injuries occur in the front of someone involved in a frontal acident.its all there you will just have to do a bit of counting

good luck lady !!!!

frontalrightleftrear243810148889695433486365161215388717113125353734sum151131118112mean18.87516.37514.7514

ta, ill give that a go. thanks

[gwh200]

im not sure about this being the right analysis for you.its contemporaneous,and requires quite a lot of jigging to make it work.
its a bit hard to explain via this forum but have a looka t this it may help
just remember you need to apply your data with regards to the inertia of each data point.if you dont get this within a few days,find a new analytical method ....quick !!

http://sru.soc.surrey.ac.uk/SRU7.html#table2