I am writing a report for my class project. I am taking statistics and I am REALLY panicking with the results I have in my report. I do not think my calculations for t-test statistic or confidence upper or lower limit/upper limit and Degrees of freedom are correct. I really need help with calculating the values correctly. The Equal column is when variances of the two samples are equal and the Unequal column is when their values are not equal.
After I change degrees of freedom for each column to 999 my values under break...
p-Value #NAME? #NAME? Null Hypoth. at 10% Significance #NAME? #NAME? Null Hypoth. at 5% Significance #NAME? #NAME? Null Hypoth. at 1% Significance #NAME? #NAME?
The degrees of freedom used in each case is different, and they also depend on the specific type of test you are doing. Without a description of the meaning of the data and how you collected it, it is difficult to determine whether your calculations are valid. If your observations for the two groups are not paired, even though the per-group sample sizes are equal, then you would use a two-sample independent $t$-test, and the degrees of freedom is calculated under the assumption of a pooled sample variance (if the two groups are assumed to have equal underlying population variances), hence it would be $n_1 + n_2 - 2 = 1000 + 1000 - 2 = 1998$; or using Satterthwaite's approximation under the assumption of unequal population variances (that formula is much more complicated so I've omitted it).
If your observations naturally pair up, then you would do a paired $t$-test and the degrees of freedom are again different. Only you know which test is appropriate because it is not possible to tell from looking at your spreadsheet what the numbers mean.
I should also take this opportunity to point out that for any of these tests you choose to do on this data, you will obtain significance with $p < .00001$ simply because the sample sizes are huge. This is because your hypothesis is of the form $$H_0 : \mu_1 = \mu_2 \quad \mathrm{vs.} \quad H_a : \mu_1 \ne \mu_2,$$ and as you can see, even a tiny difference in the true means constitutes a rejection of the null hypothesis. The true "chicken" mean could be exactly $889$ and the true "shrimp" mean could be exactly $889.001$ and given enough data, you would detect this difference even though it is not really meaningful in a practical, real-world sense. To help you understand why, if I said that a new experimental drug decreases your risk of heart attack by $0.001\%$, but it costs 100 dollars per pill, would you think it worthwhile to take? In such a context, sometimes a more meaningful statistical test of significance is to use a hypothesis of the form $$H_0 : |\mu_1 - \mu_2| \le \Delta \quad \mathrm{vs.} \quad H_a : |\mu_1 - \mu_2| > \Delta,$$ for some "clinically significant" difference $\Delta$. For instance, we might say that a drug has to reduce the risk by at least 15 percent to be considered marketable.
The point of the above discussion is to explain that your huge sample size is going to lead to rejecting $H_0$ because the structure of your hypothesis is designed to detect inequivalence to any degree, no matter how small. The sample means in your data may not seem that different, but in context of the sample size, it is. It is a little like flipping a fair coin 10 times versus 1000 times. In the former case, it would not surprise you to observe 4 heads and 6 tails. In the latter case, it would be extremely anomalous to observe 400 heads and 600 tails if the coin were fair.