# Which tests do I have to use in order to test my hypotheses?

##### New Member

I would like to test these hypotheses, but I'm still not sure which test I should use for it. Can anybody help me?

Last edited:

#### Musibrique

##### New Member
Can you please specify what type of test you are talking about? If you were referring to significance testing, then you would you use a one-tailed t-test for Hypothesis 1 since the alternative hypothesis only makes one prediction or direction within the normal distribution. As for hypothesis 2, it is actually a null-hypothesis because it says there is no effect between a male manager and an employee. If you want to test these 2 hypotheses, I believe it's best to make a null-hypothesis for Hypothesis 1 (Female no effect on employee's performance) and an alternative hypothesis for 2 (There is an effect) However, you have to test these 2 hypotheses independently.

##### New Member
Well, It is a test between subjects. I first tested this hypothesis: The presence of a relationship between manager and employee positively effects the performance evaluation of the employee.
The results of the test between subjects shows no significance, which means that the presence of a relationship has no influence on the evaluation.

But I would also like to test this seperate for men and women.

#### Musibrique

##### New Member
Well, It is a test between subjects. I first tested this hypothesis: The presence of a relationship between manager and employee positively effects the performance evaluation of the employee.
Did you included the null-hypothesis that there is no effect between a manager and an employee in your significance test?

The results of the test between subjects shows no significance, which means that the presence of a relationship has no influence on the evaluation.
I'm quite interested and concerned on whether or not you did your statistical analysis correctly. Can you please provide the sample-size, successful events, and anything that you used to calculate the p-value?

But I would also like to test this seperate for men and women.
Agree!

##### New Member
Did you included the null-hypothesis that there is no effect between a manager and an employee in your significance test?
No, I think didn't do that at all. I simply put the data in SPSS (Analyze>General Lineair Model>Univariate). With performance as the DV and relation and gender as Fixed Factors.

I have attached a file with my outputs, I hope that's what you were asking for.

Basically what I did was an experiment with to scenarios. The first one described a friend (this means that there is a relationship present) and the second a relation that is strictly professional.

I deleted the responses of participants who didn't aswered the questions correctly or failed the manipulation check.

#### Musibrique

##### New Member
No, I think didn't do that at all. I simply put the data in SPSS (Analyze>General Lineair Model>Univariate). With performance as the DV and relation and gender as Fixed Factors.
I read the data and it said, "Tests the null hypothesis that the error variance of the dependent variable is equal across groups."

I'm assuming you did included the null-hypothesis. However, my question is what is the probability or assumption of the null-hypothesis based on? For example, suppose there's a hypothesis that states that smokers get lung cancer and a null-hypothesis that smokers don't get lung cancer. We can conduct studies for this, but we can't calculate significance because it isn't really that specific. However, if the hypothesis said that smokers have a 75% chance of getting lung cancer and the null-hypothesis said that smokers don't have a 75% of getting lung cancer, then we can calculate significance because it refers to a specific population (e.g. 75%) instead of the whole population. If smokers really do have a 75% chance of getting lung cancer, then we would expect 75% of smokers in our studies to have the disease. Makes sense?

#### Musibrique

##### New Member
If you are still having trouble about what I meant on the null-hypothesis, you can google, "Introduction To Hypothesis Testing"

I hope it helps!

##### New Member
Yes! Thank you, this makes sense to me. I know my hypotheses are not specific enough but I would like to check if I understood you well based on the hypotheses I already have set up and these were:

Hypothesis 1: The presence of a relationship between a female manager and employee positively effects the performance evaluation of the employee.

Hypothesis 2: The presence of a relationship between a male manager and employee has no effect on the performance evaluation of the employee.

Can you please specify what type of test you are talking about? If you were referring to significance testing, then you would you use a one-tailed t-test for Hypothesis 1 since the alternative hypothesis only makes one prediction or direction within the normal distribution. As for hypothesis 2, it is actually a null-hypothesis because it says there is no effect between a male manager and an employee. If you want to test these 2 hypotheses, I believe it's best to make a null-hypothesis for Hypothesis 1 (Female no effect on employee's performance) and an alternative hypothesis for 2 (There is an effect) However, you have to test these 2 hypotheses independently.
I understand that my two hypotheses are not consistent with each other. So for both hypotheses the null hypothesis should be: there is no effect and the alternative hypothesis should be: there is an effect. Right?

Then, you adviced me to do a one-tailed t-test for just only women and another one-tailed t-test for just only men. In SPSS this is the independent samples T test (Analyze>Compare Means>Independent samples t test), right? Because, my DV is performance and my IV is relation.

For example, I have selected all the women in my sample and performed an independent samples T test. You can find the output in the attachment. The output shows a p-value of 0,532 (2-tailed). Based on an Alfa of 5% I can accept the null hypothesis, which was (after adjustment): The presence of a relationship between a female manager and employee has no effect on the performance evaluation of the employee.

Right?

For men, the p-value is 0,626, which also means that we should accept the null hypothesis (The presence of a relationship between a male manager and employee has no effect on the performance evaluation of the employee). Right?

So this means that gender has no effect on the performance evaluation in the presence of a relationship. Right?

There is another thing that confuses me I thought this is a 2x2 design, because:

Performance = Gender x Relation

But a friend of mine says this is a 2x1 design, based on my original three hypotheses, which are:

- Hypothesis 1: The presence of a relationship between manager and employee effects positively the performance evaluation of the employee.
- Hypothesis 2: The presence of a relationship between a female manager and employee positively effects the performance evaluation of the employee.
- Hypothesis 3: The presence of a relationship between a male manager and employee has no effect on the performance evaluation of the employee.

Is she right?

Thank you!!

#### Musibrique

##### New Member
Yes! Thank you, this makes sense to me.
You're welcome! Glad I can help

I know my hypotheses are not specific enough but I would like to check if I understood you well based on the hypotheses I already have set up and these were:

Hypothesis 1: The presence of a relationship between a female manager and employee positively effects the performance evaluation of the employee.

Hypothesis 2: The presence of a relationship between a male manager and employee has no effect on the performance evaluation of the employee.

I understand that my two hypotheses are not consistent with each other. So for both hypotheses the null hypothesis should be: there is no effect and the alternative hypothesis should be: there is an effect. Right?
Correct! A rule of thumb in statistics is to always create a null-hypothesis when doing a t-test. Otherwise, the t-test wouldn't make sense.

Then, you adviced me to do a one-tailed t-test for just only women and another one-tailed t-test for just only men. In SPSS this is the independent samples T test (Analyze>Compare Means>Independent samples t test), right? Because, my DV is performance and my IV is relation.
The reason why I advice you to do a one-tailed t-test is because the effect within those two alternative hypothesis seem to make a prediction in only one direction. I can, of course, be wrong because it is possible that the effect could go to the other direction as well. For example, a positive effect could refer to the employee doing better in its performance whereas a negative effect could refer to the employee doing bad in its performance. If positive and negative effect apply, then you should you use a two-tailed t-test. If it's either positive or negative, you should use a one-tailed t-test.

For example, I have selected all the women in my sample and performed an independent samples T test. You can find the output in the attachment. The output shows a p-value of 0,532 (2-tailed). Based on an Alfa of 5% I can accept the null hypothesis, which was (after adjustment): The presence of a relationship between a female manager and employee has no effect on the performance evaluation of the employee.

Right?
Right! Since the two-tailed p-value in your independent sample t-test is too big for the null-hypothesis to be rejected at the 0.05 significance level, you should accept the null-hypothesis. Even if we insist that the proper p-value is actually one-tailed, it would still be too big to reject the null-hypothesis:

One-tailed p-value=two-tailed p-value/2. Thus, p=0.532/2=0.26. Therefore, the one-tailed p-value is 0.26

For men, the p-value is 0,626, which also means that we should accept the null hypothesis (The presence of a relationship between a male manager and employee has no effect on the performance evaluation of the employee). Right?
Right!

So this means that gender has no effect on the performance evaluation in the presence of a relationship. Right?
Unless the statistical power in the independent sample t-test is too weak to reject the null-hypothesis, you are correct.

There is another thing that confuses me I thought this is a 2x2 design, because:

Performance = Gender x Relation

But a friend of mine says this is a 2x1 design, based on my original three hypotheses, which are:

- Hypothesis 1: The presence of a relationship between manager and employee effects positively the performance evaluation of the employee.
- Hypothesis 2: The presence of a relationship between a female manager and employee positively effects the performance evaluation of the employee.
- Hypothesis 3: The presence of a relationship between a male manager and employee has no effect on the performance evaluation of the employee.

Is she right?
Hypothesis 1 seems to refer to the whole population of managers and employees, regardless of their gender. Hypothesis 2 refers to a specific female population with employees whereas Hypothesis 3 refers to the male population as well as its employees. Since you are conducting independent t-tests rather than dependent ones, you should make the NxN design independent as well. For example, Performace= Gender X Relation apply to Hypothesis 2 and 3. In this case, you can use a 2x2 design for the females and another 2x2 design for the males. As for Hypothesis 1, you can you a 2x1 design. Hope that helps!

Last edited:

##### New Member
Yes, it helps a lot! Thank you thank you thank you!!!

##### New Member
Dear Musibrique,

I have another question. You said:

The reason why I advice you to do a one-tailed t-test is because the effect within those two alternative hypothesis seem to make a prediction in only one direction. I can, of course, be wrong because it is possible that the effect could go to the other direction as well. For example, a positive effect could refer to the employee doing better in its performance whereas a negative effect could refer to the employee doing bad in its performance. If positive and negative effect apply, then you should you use a two-tailed t-test. If it's either positive or negative, you should use a one-tailed t-test.
Assuming H0: The presence of a relationship between a female manager and employee has no effect on the performance evaluation of the employee.

H1: The presence of a relationship between a female manager and employee positively effects the performance evaluation of the employee.

Using a two-tailed t-test, because the performance could be affected positively or negetively by the presence of a relation. How do I know if this effect is negative or positive?

#### Musibrique

##### New Member
Dear Musibrique,

Assuming H0: The presence of a relationship between a female manager and employee has no effect on the performance evaluation of the employee.

H1: The presence of a relationship between a female manager and employee positively effects the performance evaluation of the employee.

Using a two-tailed t-test, because the performance could be affected positively or negetively by the presence of a relation. How do I know if this effect is negative or positive?
I'm afraid the answer to this question is entirely up to you actually. I can't say what constitutes as a negative or positive correlation because this is your own research hypothesis. If I created a research hypothesis similar to yours, my perception of what is positive and negative may be different to yours. Here is some clarity:

What would you define as a positive/negative correlation?

What correlation are you exactly looking for in your research hypothesis? Are you only looking for a positive correlation between a female manager and employee or a negative correlation or both correlations?

Hope that helps!

##### New Member
No, this makes me a bit confused.

A female manager gives her employee a rating for his/her performance. The scale of the rating is 0-100. So when there is a relationship between the female manager and employee, in this case the rating should be higher than in the case where there is a strictly professional relation. So "positive" means a higher rating of performance.

I'm looking for the correlation between the given performance rating and the status of the relation.

The Pearson's correlation is positive does this has anything to do with it?

#### Musibrique

##### New Member
Pearson's correlation basically measures the strength of the correlation between two factors. In this case, I'm assuming these two factors in Pearson's correlation are employee's performance rating and the status of the relation, correct? If so, then the correlation between the given performance rating and the status of the relation is positive. Unless, of course, you're talking about significance which is a different ball game.

#### Musibrique

##### New Member
Just to be sure, what is the value of Pearson's correlation?

##### New Member
Correct! So the correlation between employee's performance rating and the status of the relation is positive.

If you perform a t-test and the p-value is 0,532, then you accept the null hypothesis. But is the p-value is 0.001 then you should accept H1.
When you accept H1, I would say yes there the relation positively effects the performance rating, because there is a positive correlation between employee's performance rating and the status of the relation.

Or am I wrong?

##### New Member
This is just the correlation of women.

I have splitted my sample into man and women and perform the tests seperatly.

And the correlation is negative.

#### Musibrique

##### New Member
Correct! So the correlation between employee's performance rating and the status of the relation is positive.
Right!

If you perform a t-test and the p-value is 0,532, then you accept the null hypothesis. But is the p-value is 0.001 then you should accept H1.
When you accept H1, I would say yes there the relation positively effects the performance rating, because there is a positive correlation between employee's performance rating and the status of the relation.

Or am I wrong?
A p-value of 0.001 is very good statistical evidence for H1 since it has a probability of extreme or more extreme of 1,000 to 1, so you can safely reject the null-hypothesis.

As for the conclusion, you can say there is a relationship if your independent t-test is statistically significant; however, it is important to know that the relationship may not be practically significant as you think. This is one problem with t-tests. They don't tell you how strong the relationship is, only evidence. You can, however, use Pearson's correlation to determine how strong is the effect. You can also calculate the effect-size as well, but that may be too advance for you.