|
United States24701 Posts
I've never had to conduct a survey before. I'm currently taking a grad class where I will be distributing a survey to about 100-200 people, and analyzing the results. I haven't received much direction from the instructor regarding ways to tally and correlate results.
One suggestion I received was to have participants use a survey created on a website which automatically tallies results. I have two problems with this. One is that the people taking the survey won't have access to a computer when the survey is distributed, and I doubt most of them will go out of their way to complete the survey online. The second is that the primary candidate for online software I've looked up (Survey Monkey) doesn't seem to have that many options for analyzing data aside from seeing what percentage/count of respondents put a specific answer (I could be wrong about this).
Physically tallying the results isn't that difficult (although it will be tedious by hand). I could easily determine that 45% of respondents put a) for question one. However, I'm struggling with ways I could do more complex comparisons. Let me create an example to illustrate this:
Survey:
1) Do you like eggs?
a) yes
b) no
2) Do you like starcraft?
a) yes
b) no
It is easy to tell what percent of the people taking the survey like eggs, or like starcraft, but I want to figure out what correlative effect liking starcraft has on liking eggs. Maybe I'll discover that 50% of the people who don't like starcraft like eggs, whereas 90% of the people who like starcraft like eggs. How would I go about that? In principle, I'd need to separate out all of the survey results that responded 'yes' to question 2 and then do a new tally of question 1 (and vice versa). But if I were tallying by hand and there were a couple of dozen questions, it would be a very large amount of work to search through for different correlations.
Is there software I could use where I plug all the data from each survey in, and I then have the flexibility to do any analysis I want? Are there other techniques I could use? I was thinking maybe it's reasonably possible to use an Excel spreadsheet where I can sort results... but I've never tried something like that before for this application so I'm not sure how viable it is.
If you have a good idea or have experience in this area, please let me know. Also, if you can refer me to a source where I can learn more about this (that's readily available preferably) then please let me know. I'm actually jumping the gun a bit with worrying about how I am going to analyze the data before I've even finalized my methods of collecting it, but I'm curious what's available.
  
|
excel will be fine. simple pivot tables will probably do everything you're looking for. or i guess you can also filter columns and look at the counts.
there are definitely better programs out there for working with large datasets but i've only worked with them in a corporate environment and i imagine they cost a couple of thousand dollars which probably isn't what you're looking for (plus the dataset here is tiny anyway)
edit: well to be fair, excel is still better in my experience. it's just that it's limited to a certain number of records which isn't an issue here
|
Stata has everything you need.
|
Thers like 999999 different ways to find a correlation
For instance, to show correlation, lets say 60% of people like eggs. Yet 90% of people who like starcraft like eggs. The expected number would be 60%, so use a chisquared test with two degrees of freedom to get a p value which would suggest lack of correlation or correlation.
|
United States24701 Posts
On March 01 2009 05:05 Caller wrote:
Thers like 999999 different ways to find a correlation
For instance, to show correlation, lets say 60% of people like eggs. Yet 90% of people who like starcraft like eggs. The expected number would be 60%, so use a chisquared test with two degrees of freedom to get a p value which would suggest lack of correlation or correlation.
This has nothing to do with this blog. If I need help or want input when I'm doing the actual math, then sure we can discuss it then.
|
United States24701 Posts
On March 01 2009 04:56 defenestrate wrote:
Stata has everything you need.
Okay I looked this up. It's $665 dollars if you buy the small version for a single license. Obviously this is unreasonable for the purposes of this assignment, but do you know of any alternatives? I guess I could ask my university if they have access to it but I doubt it.
|
Hmm, in research i always used the root framework (see root.cern.ch) for data analysis purposes. It however is more a tool for complex analyses like when you have thosands or millions of datasamples and in addition might take quite some time to get into.
|
United States24701 Posts
On March 01 2009 06:30 Thojorin wrote:
Hmm, in research i always used the root framework (see root.cern.ch) for data analysis purposes. It however is more a tool for complex analyses like when you have thosands or millions of datasamples and in addition might take quite some time to get into.
Should I even bother looking at that without a decent amount of programming experience, or would it be too difficult to use?
|
On March 01 2009 07:05 micronesia wrote:Show nested quote +On March 01 2009 06:30 Thojorin wrote:
Hmm, in research i always used the root framework (see root.cern.ch) for data analysis purposes. It however is more a tool for complex analyses like when you have thosands or millions of datasamples and in addition might take quite some time to get into. Should I even bother looking at that without a decent amount of programming experience, or would it be too difficult to use?
Well, that depends on the time you are willing to spend. I think you could get the desired result within some days or so, which mainly consts of browsing through the software documentation. It may take longer if you have no programming experience whatsoever.
However, i am not aware of any "simple" spftware doing the job for you. Sorry about that.
|
United States24701 Posts
On March 01 2009 07:23 Thojorin wrote:Show nested quote +On March 01 2009 07:05 micronesia wrote:On March 01 2009 06:30 Thojorin wrote:
Hmm, in research i always used the root framework (see root.cern.ch) for data analysis purposes. It however is more a tool for complex analyses like when you have thosands or millions of datasamples and in addition might take quite some time to get into. Should I even bother looking at that without a decent amount of programming experience, or would it be too difficult to use? Well, that depends on the time you are willing to spend. I think you could get the desired result within some days or so, which mainly consts of browsing through the software documentation. It may take longer if you have no programming experience whatsoever. However, i am not aware of any "simple" spftware doing the job for you. Sorry about that.
I was considering Database software (Access) and/or excel to do the job. I am trying to figure that out now.
|
Actually most universities should have access to Stata, it is one of the more common tools for stat classes far as I know. Statool is one that some friends have been using for a while as it is free, though I can't attest for the quality myself.
|
United States22883 Posts
Write your questions carefully and it might be a good idea to run it by us before you give it out. What kinds of questions are you actually going to ask? I'm sure you realize yes/no is too simple for most things, but also make sure you don't use direct rating scales. Rankings are fine, but for legitimate survey work you want people comparing items (starcraft or eggs?) rather than saying "eggs are a 4, starcraft is a 5." It ends up being a lot more work as you increase the options, but you get shitty data with scales because 1. you're mixing cardinal and ordinal numbers so everything gets misrepresented and 2. person x's 10 may be less favorable than person y's 10.
Stata or even Excel should do what you want. SurveyMonkey is good for written responses and improvement questionnaires, not so much for quantitative research.
|
United States24701 Posts
On March 01 2009 09:30 Jibba wrote:
Write your questions carefully and it might be a good idea to run it by us before you give it out. What kinds of questions are you actually going to ask? I'm sure you realize yes/no is too simple for most things, but also make sure you don't use direct rating scales. Rankings are fine, but for legitimate survey work you want people comparing items (starcraft or eggs?) rather than saying "eggs are a 4, starcraft is a 5." It ends up being a lot more work as you increase the options, but you get shitty data with scales because 1. you're mixing cardinal and ordinal numbers so everything gets misrepresented and 2. person x's 10 may be less favorable than person y's 10.
Stata or even Excel should do what you want. SurveyMonkey is good for written responses and improvement questionnaires, not so much for quantitative research.
I'm actually not going to release the survey, but I can tell you that it's mostly Likert scale questions with 1 being strongly agree and 5 being strongly disagree. There will be a couple of multiple choice questions afterward with an 'other' blank to write in an answer. I've already run it by the professor/classmates, and I am currently editing it.
Yeah my example was not very good for legit research... but I think it illustrated the technical difficulties I was worried about to those who didn't get what I meant.
If my university doesn't have Stata available is there any legit way for me to use it? I really can't afford to pirate something for this since my work could ultimately get published.
|
We used SPSS(google it) at school.
Though the interface is a lil old school, it has all the functionality you need.
For stuff like this we used to do :
Correlations, Calculating Reliability, Linear Approximation, Statistical Significance
|
United States22883 Posts
Are you asking the same questions in multiple forms? It gets annoying for the survey taker, but it's good for you.
|
Could you get them to fill out a scantron sheet and just run it through the machine that automatically reads them?
Then just load it into matlab and do some data mining.
edit: maybe you do want some software to analyze it for you....it could get complicated
|
United States24701 Posts
On March 01 2009 09:41 LastWish wrote:
We used SPSS(google it) at school.
Though the interface is a lil old school, it has all the functionality you need.
For stuff like this we used to do :
Correlations, Calculating Reliability, Linear Approximation, Statistical Significance
Ah I recall my professor mentioning it. I'll check if I can get this one somehow.
On March 01 2009 09:43 Jibba wrote:
Are you asking the same questions in multiple forms? It gets annoying for the survey taker, but it's good for you.
Can you explain this? What is the rationale. I'll check when I edit my survey to see if I already do this, and to what extent. However, I want to make the survey short and sweet for many reasons.
On March 01 2009 09:49 fight_or_flight wrote:
Could you get them to fill out a scantron sheet and just run it through the machine that automatically reads them?
Then just load it into matlab and do some data mining.
I thought about this, but I am not aware of any technology that I have easy access to that allows me to load data into a computer. I have access to a scantron where I work but all it does is tabulate a few overall statistics... and I'd get into trouble for wasting all their forms anyway XD
|
You could write a GUI based program on a computer and get people to take that as their survey.
|
United States24701 Posts
On March 01 2009 10:00 fight_or_flight wrote:
You could write a GUI based program on a computer and get people to take that as their survey.
"One suggestion I received was to have participants use a survey created on a website which automatically tallies results. I have two problems with this. One is that the people taking the survey won't have access to a computer when the survey is distributed, and I doubt most of them will go out of their way to complete the survey online." -OP
I really would like to do something like this but I won't be able to :p
|
|
|
United States24701 Posts
On March 01 2009 10:07 fight_or_flight wrote:
I mean on a laptop.
Do you mean have one person at a time use the laptop? I wouldn't be able to bring multiple laptops and one would take too long.
|
|
|
dunno, if you're lookin for statistical software that does all calculations for you and spits them out minitab is good. They give a free 60 day trial so that's good, and if you're not to shy about it, there's a cracked version of it for free.
|
From a programming standpoint, it would be somewhat like having four variables.
if(sc && eggs)
scAndEggs++;
etc. etc.
Though I'm assuming you need significantly more complicated statistics. I'd suggest relying on a program, but if you need something relatively simple, I'm sure some people could write a quick and dirty one for you.
|
United States24701 Posts
I'm glad you guys have made these suggestions, but they are going to be fallbacks in case my current plan fails.
Someone was showing me how you could use Microsoft Access to get the kinds of statistics I want. The answer to each question on the survey is a field, and then I create a query which rejects all the entries which don't have fields that meet the requirements. Then I can see how many are leftover. The trickier part is dealing with the results of the Likert questions where the average shifts from 3.5 to 4.2 and stuff like that.... but he claims you could either do it in access or just transfer the data over to excel to do the calculation once access has organized it.
|
Belgium8305 Posts
i think either surveymonkey or surveygizmo lets you export the results to an excel file
either way, when you're dealing with 100-200 respondents, it's probably best to manually feed the results into SPSS or whatever, from there you can run any analysis you want
it's tedious and a shitty job, but you could get some people to help you for a small compensation and it'll take just a few hours probably
|
|
|
|
|
|
EPT
