You haven't described what you're trying to accomplish or test.
Hello everyone,
I am not so familiar with statistics but I have a question concerning the computation of a variable.
I have collected data on certain crowdfunding offerings. I scanned the initiaitves and checked it for the following (usage of twitter, facebook, youtube and updates) and coded it the following:
Twitter (0=No, 1=Yes)
Facebook (0=No, 1=Yes)
Youtube (0=No, 1=Yes)
Updates (0=No, 1=Yes)
But I don't want to test, how they are using one particular channel. I would like to compute a Social Media index. How? Simply aggregating the data on these channels.
Campaign A uses twitter and facebook scores a 2 then.
Is such a thing permissible?
Thank you very much. I highly appreciate it!
You haven't described what you're trying to accomplish or test.
You could create a new variable in the following way: Number of social medias (0, 1, 2, 3, 4), and then use a binary logistic regression with the funded variable as the dependent variable. Or you can run a logit regression with the four variables you now got to see which contribute most to the funded variable.
Sounds like a probit would be good with the binary Funded variable as the dependent variable.
I'm not knowledgeable enough to know whether it'd be better to construct an index or to include the 4 social media variables as separate regressors. Multicollinearity would be one of the necessary considerations.
Well, I don't need to go into any detail concering which social media channel contributes most and so on. That's why I thought a computation like this is enough.
What do you think?
In that case make the index like you originally proposed, and use either a probit or logit regression with Funded as the dependent variable.
hey derksheng,
thanks for coming back so quick. Why can't I use the Social Media index as the dependent and the funded as the independent and compute a independent t-test?
Cheers
Probit/Logit are nice and can only be done with binary dependent variables. It also has the interpretation of "given X social media usage, we expect them to be funded Y % of the time".
@derksheng
I don't get it. When I compute the Social Media Index, I will receive a variable that won't be binary (0;1) any more. It'll be 0-4. How does this then look like?!
Thank you!
julia89 (07-17-2012)
julia89 (07-17-2012)
If you think you've come up with a betting plan that turns a negative expected turnover into a positive, then drop that thought!
That's closer but still not quite right (the error isn't inside there).
Typically you say something like
(which is equivalent to)
In this case Y is Funded and X is Index. N doesn't have to be 1 (we could replace 1 with but typically logistic regression is done on bernoulli trials).
The fun thing about generalized linear models (which logistic regression and probit regression are a part of) is that you get to get out of the "signal + noise" type of mindset where you specify an expected value and then some error on top of that. Generalized linear models make you think in terms of a response distribution conditional on the covariates.
I don't have emotions and sometimes that makes me very sad.
Englund (07-16-2012)
Tweet |