hypotheses help needed and excel/spss q!

alyk

hi,

im currently in the midst of my MSc thesis and was wondering if anyone had any advice on how to formulate hypotheses?

in the past i have done research that meant i formulated hypotheses from the research that i was aiming to improve on by accounting for their limitations etc. however, this time round im looking at variables that are not researched together previously (or in a limited way). any suggestions welcome?!

also, is there any easy way of transfering data from excel to spss?

thanks

2Scoops

alyk wrote: »

in the past i have done research that meant i formulated hypotheses from the research that i was aiming to improve on by accounting for their limitations etc. however, this time round im looking at variables that are not researched together previously (or in a limited way). any suggestions welcome?!

Well, it's as simple as 'what do you think will happen?' Support it with as much relevant background as you can showing it is a reasonable hypothesis with important implications.

alyk wrote: »

also, is there any easy way of transfering data from excel to spss?

Couldn't be easier - SPSS will open Excel files directly.

Ostrom

In this case they are usually arrived at theoretically / inductively i.e. how well particular combinations address your overall research question. If you haven't stated clear predictions at earlier stages then perhaps run a number of models and check for those offering greatest explanatory potential? Unless your literature is specific on which variables should be introduced/tested in particular ways?

Can also highlight the excel data and copy and past into data view in SPSS if you wish.

alyk

thanks to those that replied, it was very useful!!:D i now have my hypotheses sorted!

i have a few questions before i can start my analysis that im stuck on, so any suggestions would be welcome

1) im looking at personality as a predictor, it has 7 dimensions/facets. how do i compute a total personality score? do i just compute a new variable and add all 7 dimension total scores?

2). anyone have experience with moderated Mutliple regression? i know how to do the analysis, but looking for any reccomendations of websites/ books that help with the interpretation of the output and reporting the results?had to do moderated MR for an assignment and these were the two areas that effected my mark, so wanna get it right for my dissertation

finally, im aware of the process of dealing with missing values in the data set, however, is there a way to code/deal with missing demographic information?(not including age). so all are non numerical.

ANY suggestions would be most appreciated!!

Ostrom

alyk wrote: »

thanks to those that replied, it was very useful!!:D i now have my hypotheses sorted!

i have a few questions before i can start my analysis that im stuck on, so any suggestions would be welcome

1) im looking at personality as a predictor, it has 7 dimensions/facets. how do i compute a total personality score? do i just compute a new variable and add all 7 dimension total scores?

You can calculate the new variable by using the [transform -- compute] dialogue (syntax is similar to excel), and produce either an average or summed score. Is each dimension measured with a certain number of variables? If so you should also run some reliability checks under analyse --- scale --- reliability analysis. Criteria for alpha levels vary by discipline and by measure so maybe check the literature for satisfactory reliability scores.

If you are using a number of individual personality dimensions as separate predictors watch out for colinearity.

alyk wrote: »

2). anyone have experience with moderated Mutliple regression? i know how to do the analysis, but looking for any reccomendations of websites/ books that help with the interpretation of the output and reporting the results?had to do moderated MR for an assignment and these were the two areas that effected my mark, so wanna get it right for my dissertation

Might be a confusion of terms on my part - are you talking about computing an interaction term? akaik you can compute the variable yourself with the same dialog as above, but get a second opinion please!

Is the moderator categorical?

alyk wrote: »

finally, im aware of the process of dealing with missing values in the data set, however, is there a way to code/deal with missing demographic information?(not including age). so all are non numerical.

ANY suggestions would be most appreciated!!

Not sure what you're getting at - you want your demographic variables to be categorical (non-numerical)? Usual (quick fix) approach would be mean substitution for missing quant data. You can recode age into categories afterwards if you would rather work that way but this would not be suitable for regression if you already have the detail in the interval score. Sage have a specific book in their quantitative methods series dealing with missing data that might be worth a look.

http://books.google.com/books/about/Missing_data.html?id=ZtYArHXjpB8C

alyk

efla wrote: »

You can calculate the new variable by using the [transform -- compute] dialogue (syntax is similar to excel), and produce either an average or summed score. Is each dimension measured with a certain number of variables? If so you should also run some reliability checks under analyse --- scale --- reliability analysis. Criteria for alpha levels vary by discipline and by measure so maybe check the literature for satisfactory reliability scores.

If you are using a number of individual personality dimensions as separate predictors watch out for colinearity.



Might be a confusion of terms on my part - are you talking about computing an interaction term? akaik you can compute the variable yourself with the same dialog as above, but get a second opinion please!

Is the moderator categorical?



Not sure what you're getting at - you want your demographic variables to be categorical (non-numerical)? Usual (quick fix) approach would be mean substitution for missing quant data. You can recode age into categories afterwards if you would rather work that way but this would not be suitable for regression if you already have the detail in the interval score. Sage have a specific book in their quantitative methods series dealing with missing data that might be worth a look.

http://books.google.com/books/about/Missing_data.html?id=ZtYArHXjpB8C

i have realised since that there can be no total personality score, so now looking at the 7 dimenesions and have checked collinearity in statistics analysis. is there a way to deal with minor collinearity, if say VIF values are 1.000 or slightly above?
all my varaibles are continious (not using demographics in main analysis). what is confusing me is the idea of when u need to center a variable?the regression analysis im now using, linear, standard multiple regression and hierarchal. some say always center continious predictors for regression and others are saying only if looking at interaction terms, for mediated or moderated regression? has me a bit confused?!

Ostrom

alyk wrote: »

i have realised since that there can be no total personality score, so now looking at the 7 dimenesions and have checked collinearity in statistics analysis. is there a way to deal with minor collinearity, if say VIF values are 1.000 or slightly above?
all my varaibles are continious (not using demographics in main analysis). what is confusing me is the idea of when u need to center a variable?the regression analysis im now using, linear, standard multiple regression and hierarchal. some say always center continious predictors for regression and others are saying only if looking at interaction terms, for mediated or moderated regression? has me a bit confused?!

How far from 1 is it? There are no really specific rules for interpreting VIF - aside from the broad limits - (i.e. many papers and books will work with different levels). You should also be working from average VIF afaik. Did you check the distribution of variance proportions across eigenvalues? This might give you a bit more scope for argument in your write-up as this diagnostic doesn't work with specified critical values (i.e. you just eyeball the distribution for loading). You might need to check an additional box in SPSS to get this table if you dont have it already.

I dont think there is much you can do in the way of transformation to avoid this - I'm just looking at Andy Field's book which states you can use factor analysis scores if there are a number of variables producing colinearity, but this sounds a bit extreme. Can you drop any of the predictors?

I'm running out of expertise on this as I'm not sure what you can and cannot do with these measures in psychology (I'm just a lowly sociologist!)

alyk

efla wrote: »

How far from 1 is it? There are no really specific rules for interpreting VIF - aside from the broad limits - (i.e. many papers and books will work with different levels). You should also be working from average VIF afaik. Did you check the distribution of variance proportions across eigenvalues? This might give you a bit more scope for argument in your write-up as this diagnostic doesn't work with specified critical values (i.e. you just eyeball the distribution for loading). You might need to check an additional box in SPSS to get this table if you dont have it already.

I dont think there is much you can do in the way of transformation to avoid this - I'm just looking at Andy Field's book which states you can use factor analysis scores if there are a number of variables producing colinearity, but this sounds a bit extreme. Can you drop any of the predictors?

thanks for your reply!no the VIF values are mainly 1 with some marginally above 1 e. g. 1013. lost at the description of using loadings of the distribution and factor score as im not using factor analysis.
all my analysis are correlations and then multiple regressions. so what im trying to work out is if/when u need to center continious predictors?i'ave also read they help with collinearity.
in relation to predictors, if they are not significant in the correlation, then im dropping them for the regression analysis.

Ostrom

alyk wrote: »

efla wrote: »

thanks for your reply!no the VIF values are mainly 1 with some marginally above 1 e. g. 1013. lost at the description of using loadings of the distribution and factor score as im not using factor analysis.
all my analysis are correlations and then multiple regressions. so what im trying to work out is if/when u need to center continious predictors?i'ave also read they help with collinearity.
in relation to predictors, if they are not significant in the correlation, then im dropping them for the regression analysis.

The loadings are from your regression diagnostics - these are different from the ones you get with the factor analysis readout, it is another way of checking colinearity