# Need help urgently

#### Kate10

##### New Member
I need to run a test to run if a relationship is significant.
Hyphothesis
H1: Entertainment of the advertisement on web is positively associated with the consumers’ attitude toward the advertising

How do I develop a null hyphothesis and test which one will be accepted? Which test to use? My professor said I need to present and discuss the hypotheses using the null and alternative statements. Could someone help me? I just got the news and deadline is so soon

Thanks.

#### Mean Joe

##### TS Contributor

Null hypothesis aka H0: Consumers' attitude toward the advertising has no association with their entertainment.

As for the test you use...You could do a chi square test. This would entail asking a couple of two possibility questions: 1) What is your attitude toward the advertising? Favorable/Unfavorable. 2) Were you entertained by the ad? Yes/No.

There's other possibilities as well, but you didn't give us much background about what class this is for. Since you mentioned a professor, I'll infer it's a college class, but there's a big difference between the expectations of a 100-level college class, and a graduate thesis. The chi square test is a 100-level possibility.

#### Kate10

##### New Member
Hi Mean Joe,

Thanks for your reply. I actually noticed how poor my post was my self. I just paniced of my professors comments just before deadline.

So more details.

This is my question: To what extent do you agree or disagree on the following statements about advertising being entertaining.

1. strongly agree 2. agree 3. neither 4. disagree 5. strongly disagree

ENTERTAINMENT
(All these variables have been computed as a single variable)

Other question is:
1. strongly agree 2. agree 3. neither 4. disagree 5. strongly disagree

Over all attitude towards advertising is positive

I have used Spearmans' rank correlation to test the correlation between entertainment and attitude towards advertising. I have the p value is that enough to accept or reject null hypothesis or do i need the t-value?

I hope this clarified my problem..Statistics are not really my strenght and my supervisor seems to be even worse than me. He gives opinions but doesn't know how to do that...:/

#### CB

##### Super Moderator
I have used Spearmans' rank correlation to test the correlation between entertainment and attitude towards advertising. I have the p value is that enough to accept or reject null hypothesis or do i need the t-value?
Sounds a sensible enough approach. If you have the p value you don't need the t-value for decision-making purposes - the t-statistic value is just used to calculate the p value. If your p value is less than your chosen alpha level (e.g. 0.05) you can reject the null hypothesis.

#### Chloe Paris

##### New Member
help: simple correlation with a 3x3 contingency table?

Hi there

I might have very similar data: I would like to test students' knowledge of foreign language and especially to find out what factors help the students improve this knowledge.
So I have two factors (indep variables) - a) time spent abroad in the target language country b) time spent learning the language in classroom.

I gave them a questionnaire and they had several questions with 3 possible answers. The replies were rated as a) good b) approximately good c) wrong.
So this is considered as the dependent variable with 3 categories.

I would like to know which test is appropriate for this kind of data.
I thought the simple correlation (with a 3x3 chi-square contingency table) would be the best for me - so I want to use it but I dont know if its
the best. I need to hand in my dissertation in sept so I dont have much time to learn anything new or use any other complicated test, seeing that i've

Can anyone tell me if I'm on the right track please ??? thanks a lot!

ps: example of data

(Independent variable: Time abroad; Dependent variable: Translation of a word)

Good translation x x x
Approximate translation x x x
Wrong translation x x x

Basically I want to argue that time spent abroad has a more significant effect
on their knowledge, than the time spent learning the language in classroom.
Is this a good test? thaaanks!!!

#### CB

##### Super Moderator
Hi Chloe :welcome:

1. Are you hoping to assess the effect of your IV's on each DV item (e.g. translating a specific word) separately, or are you hoping to combine your DV questions into an overall index of foreign language knowledge? How many DV questions were there?
2. I notice you've summarised your time-abroad IV into just 3 categories. Is this how the variable was originally measured, or have you collapsed a more specific measurement scale into these 3 categories?
3. Are you just wanting to assess the individual relationships of each IV with the DV(s), or are you wanting to assess the unique predictive capacity of each IV when entered into a predictive model together?

I need to hand in my dissertation in sept so I dont have much time to learn anything new or use any other complicated test, seeing that i've
I don't mean to pick on you as I'm sure we've all been in a similar situation, but I think this is a good reminder of why it's important to ask for statistical/methodological help early in a study - preferably during the design phase! Given that you've already designed your study, collected your data, and understandably don't have time to learn any particularly complex or unfamiliar techniques, the breadth of help a statistical advisor can provide is unfortunately quite limited. Bit of a lesson for us all maybe! Hopefully we'll be able to come up with a little bit of advice anyway.

#### Chloe Paris

##### New Member
Hi Cowboy bear

1.) yes, each section in the questionnairethe has a few questions. Lets say, the 'translation' section has four questions asking students to translate a word. Since they are all very similar informal words, I collapsed them into one variable ('variable translation'; a dependent variable).

2.) I want to see whether there is a correlation between this variable and two independent variables (dealt with separately):

a) variable 'time spent learning Spanish in classroom context'
(dividing students into 3 categories, 1: 0-7 years, 2: 8-14 years, 3: more than 14 years, so I end up with just 3 categories)

b) variable 'time spent abroad - in native context' (again: dividing students into 3 categories, 1: 0-6 months, 2: 6-12 months, 3: 13months +more)

First, I dont know if I should treat all variables as categorical, since I converted them into clear categories? (or should they be ordinal, or interval?)

Initially I made several 3x3 contingency tables and ended up with chi-square values and significance values which seemed to indicate that independent variable (b) is significant (of course) ... that's basically what I wanted to show.

Maybe some other test is better for this problem, but I cant use any other test, contingency table is the only one I can master - so I basically want to justify it's the good one

Thanks for any kind of help!
C.

#### Chloe Paris

##### New Member
oh and, if it's not very clear, the replies students gave me for all the translation tasks were collapsed into just three categories again:

a) good translation b) approximate translation c) wrong translation.

Then I cross tabulated these with one independent variable, then the other.

Cheers :wave:

#### CB

##### Super Moderator
1.) yes, each section in the questionnairethe has a few questions. Lets say, the 'translation' section has four questions asking students to translate a word. Since they are all very similar informal words, I collapsed them into one variable ('variable translation'; a dependent variable).

...oh and, if it's not very clear, the replies students gave me for all the translation tasks were collapsed into just three categories again
Hi again Chloe, glad to help if I can. I must say though that I'm confused as to why you have collapsed both your IV's and DV's so severely. Categorising continuous variables into a small number of ordered categories reduces your sensitivity and reliability of measurement, limits the analyses you can do, and is likely to reduce the relationships between IV's and DV's. If you can go back to working with the uncollapsed variables (i.e. summated total of individual question scores for each DV, actual number of months or years for the IV's) this would be a better approach, and in fact given a couple of assumptions could allow you to just use a conventional correlation analysis.

First, I dont know if I should treat all variables as categorical, since I converted them into clear categories? (or should they be ordinal, or interval?)
They're currently ordinal (or "ordered-categorical"). The original measurement scales would arguably have been interval (although for the DV's this would have been a bit ambiguous, and treating them as interval could have only been appropriate once item scores were summed across a DV category).

Initially I made several 3x3 contingency tables and ended up with chi-square values and significance values which seemed to indicate that independent variable (b) is significant (of course) ... that's basically what I wanted to show.
I'm not actually sure that this is what you wanted to show The significant chi-square simply indicates that the proportion of students falling into the good/approx good/wrong answer categories is different in students with different amounts of time spent abroad. It doesn't tell you the strength or, crucially, the direction of the association; so it doesn't directly give you justification to claim that foreign language knowledge is better in students with more time abroad. Because your variables are currently ordinal, a rank-based correlation would seem most appropriate - probably Kendall's tau, or perhaps Spearman's rank-class correlation coefficient. Both of these are simply to run and available in most statistical packages.

#### Chloe Paris

##### New Member

Hi again!

thanks so much for your answer, finally someone who's told me what I'm doing wrong! Plus what you said really makes sense, gosh, this is what it looks like when someone who's never learnt stats is trying to use it :-/

I guess I just used a contingency T because it seemed the easiest
and most logical option, but in my case I see why it would reduce sensitivity of the analysis.

Well I'm going to read up on that Kendall Tau and look for some kind
of software that does it. (If by any chance you're a Mac user and you
know of a nice and user-friendly software, could you pls recommend?)

:tup:

#### Chloe Paris

##### New Member
Dear CowboyBear

sorry to bother you again, but Ive just noticed one little nuance. The nature of the questionnaire items is not the same. So with the translation, the DV is a three-point one:

a) good translation b) approximate translation c) wrong/missing translation

However I also ask students if they use the word in question, and I also want to test this "use" statistically. Then the DV is "use of word", which is a two-point one:

a) yes or b) no

Then I want to correlate these dependent variables separately with one independent variable (time abroad) and then the other (time in classroom).

Do I use the rank correlation test (e.g. Spearman's) for both of these problems? Since the first DV is categorical and not ordinal (but the IVs both ordinal, as you said)...

The other thing is that I'm not exactly sure how to code these things seeing that I actually investigated the knowledge and use of 4 words; I guess I could use a program like Goldvarb but no no no - too difficult).

I guess I could just code it all in Excel and code the same student four
times (since I tested four words). So let's say if a student spent 15 months abroad and uses 3 words but not the fourth one, I would code as follows (where 1 is "yes" and 2 is "no"):
15- 1
15- 1
15- 1
15- 2

Would this be the way to do it, and then run the data in an SPSS in Spearman's test?

Thanks for ur help!!!!! :yup:

#### CB

##### Super Moderator
Well I'm going to read up on that Kendall Tau and look for some kind
of software that does it. (If by any chance you're a Mac user and you
know of a nice and user-friendly software, could you pls recommend?)

:tup:
Hi again

I'm not a Mac user, but I know SPSS comes in a Mac version if you're looking for a complete package for social science stats, and has Spearmans/Kendalls/Pearson correlation options. There is an online calculator for Kendall's tau here which is a decidedly cheaper (free) option.

However I also ask students if they use the word in question, and I also want to test this "use" statistically. Then the DV is "use of word", which is a two-point one:

a) yes or b) no

Then I want to correlate these dependent variables separately with one independent variable (time abroad) and then the other (time in classroom).
Hmm. You know, you could arguably still regard the use-of-word variable as ordinal; the answer "yes" reflects more use of the word than "no" (obviously!) So the measurement isn't purely nominal; it has an ordinal quality. You might therefore still potentially use a correlation, though it might make a little more sense to just use a rank-based differences test (i.e. the Mann-Whitney test, which would let you see if there is a significant difference in the ranks of IV's between participants answering "yes" and participants answering "no".

The other thing is that I'm not exactly sure how to code these things seeing that I actually investigated the knowledge and use of 4 words; I guess I could use a program like Goldvarb but no no no - too difficult).

I guess I could just code it all in Excel and code the same student four
times (since I tested four words). So let's say if a student spent 15 months abroad and uses 3 words but not the fourth one, I would code as follows (where 1 is "yes" and 2 is "no"):
15- 1
15- 1
15- 1
15- 2
I'm assuming you're just wanting to treat use of each word as an individual DV rather than combine responses across the 4 words (?) In SPSS the convention would usually be to have participants as rows and variables as columns. Months abroad would be a single variable, and then each word would be a variable too with responses dummy-coded as you suggest.

#### Dason

Hi again

I'm not a Mac user, but I know SPSS comes in a Mac version if you're looking for a complete package for social science stats, and has Spearmans/Kendalls/Pearson correlation options. There is an online calculator for Kendall's tau here which is a decidedly cheaper (free) option.
Or you could go the free route and get R. You have the option of getting any of Spearman/Kendall/Pearson correlation using the base
Code:
cor(x, y, method = "spearman")
cor(x, y, method = "kendall")
cor(x, y, method = "pearson")
R is free and really quite powerful. It's available on all platforms I know of and the basic GUI for Mac is much better than the windows counterpart.

#### Chloe Paris

##### New Member
Thanks very very much guys, this is extremely helpful!

I will just code the data in Excel first, as most of the SPSSs programs seem compatible with it so I can test several of them later on. Ive dowloaded SOFA for instance, its got a very nice icon when its open - that already is a good sigh :yup:

Thanks also for suggesting "R" ... the guy who developped this software had a talk at our uni the other day (presenting R) but at this stage its a bit too complex for me ... I'm not very good at these things, to say the least.

Oh, by the way, I wondered what the "CORREL" function in Excel stands for,
do you happen to know if it's a stats test (and if so, is it pearson's, spearman's or k.tau ?)

Have a lovely weekend
Cowboy bear & Dason