# What test to perform on these hypothesis?

#### Maxime

##### New Member
Dear all,

I am having trouble trying to find the right test for my research data, and I was wondering if somebody could help me out!
These are the hypothesis:

1. The development of any proprietary technology, processes or procedures by a new firm is likely to positively affect the number of employees of that firm over time.
2. The development of any proprietary technology, processes or procedures by a new firm is likely to positively affect the business value of that firm over time.
3. A positive attitude toward innovation in a new firm is likely to positively affect the number of employees of that firm over time.
4. A positive attitude toward innovation in a new firm is likely to positively affect the business value of that firm over time.

H1: Independent = Binary variable (yes/no) Dependent = Ratio, discreet (# of employees)
H2: Independent = Binary variable (yes/no) Dependent = Ratio, continous ($business value) H3: Independent = Ordinal scale (5-point scale) Dependent = Ratio, discreet (# of employees) H4: Independent = Ordinal scale (5-point scale) Dependent = Ratio, continous ($ business value)

I am getting confused deciding which test I should use because of the binary and the ordinal scale variables. I'm guessing it's got to be some non-parametric tests, chi-square for H1 & H2 maybe?

#### trinker

##### ggplot2orBust
There are numerous ways to approach this and any one test isn't necessary right. George E. P. Box once wrote: "essentially, all models are wrong, but some are useful". This really captures the importance of researcher choice in representing the data.

I am getting confused deciding which test I should use because of the binary and the ordinal scale variables. I'm guessing it's got to be some non-parametric tests, chi-square for H1 & H2 maybe?
Think of a binary variable as a numeric way of representing categorical data.

Here are some ways I may approach your research questions:
H1: Independent = Binary variable (yes/no) Dependent = Ratio, discreet (# of employees)
number of employees suggests a Poisson or negative binomial distribution though a t-test or anova is likely robust enough to handle the data.

H2: Independent = Binary variable (yes/no) Dependent = Ratio, continous ($business value) Likely a t-test/anova approach works here as you have two groups (yes/no) with a continuous outcome H3: Independent = Ordinal scale (5-point scale) Dependent = Ratio, discreet (# of employees) Again, number of employees suggests a Poisson or negative binomial distribution though an anova is likely robust enough to handle the data H4: Independent = Ordinal scale (5-point scale) Dependent = Ratio, continous ($ business value)
An anova is likely how I would approach this with the 5 points as 5 different groups

PS excellent work laying out your question by keeping it succinct while providing the pertinent information.

#### Maxime

##### New Member
Your help is highly appreciated and I your answers somehow make sense when I think of it, but I've got a few things that I do not understand, and I didn't give you all the info needed to give the right answers I guess;
Questions:

• - 1. A t-test/anova/(negative)binomial is for testing the equality of means right? But I want to see if there is any relationship between the variables, so I was thinking of a kind of correlation model for each hypothesis?!

• - 2. I have given insufficient info I think: There was a questionaire (from the PSED II database) for companies that had to be filled in every year. So the number of employees and business value variables are meausured EVERY YEAR, as a different variable in the dataset, and so is the binary variable. So I have to test the positive relationship with the dependent variable OVER TIME! How do I do this? I think (as you mentioned) with a Poisson model, but how exactly? Poisson regression doesn't seem to be right to me because the binary variable as independent can never give a guess about the dependent variable since it's just a 1.00 or a 0.00!?

Thank you so much Trinker! You are the hero of my research

Last edited:

#### Maxime

##### New Member
Does anybody have an idea about how to handle these data that are measured repeatedly over time?

#### Maxime

##### New Member
I am really sorry to act like a spammer, but it is quite urgent! If there is anybody out there that got an answer to my questions I am so grateful!

#### noetsi

##### No cake for spunky
There are many ways to handle data repeatedly over time. Repeated measure ANOVA, time series, HLM or SEM for example. It depends on how your dv is coded, what data you have and your specific question. For example ANOVA won't, as far as I know handle a bivariate DV. Or at least the common forms won't.

Which methods have you worked with? What form will your data take.