# Need help with statistics

**URL:** <https://boards.straightdope.com/t/need-help-with-statistics/429289>\
**Category:** Factual Questions\
**Created:** [December 8, 2007, 4:10pm UTC](https://boards.straightdope.com/t/need-help-with-statistics/429289 "2007-12-08T16:10:11Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Vlad\_Igor](https://avatars.discourse-cdn.com/v4/letter/v/7feea3/32.png) [@Vlad\_Igor](https://boards.straightdope.com/u/Vlad_Igor)\
**Post date:** [December 8, 2007, 4:10pm UTC](https://boards.straightdope.com/t/need-help-with-statistics/429289/1 "2007-12-08T16:10:11Z")

</div>

…specifically the t-test. My dissertation project has been to expose cancer cells to a range of concentrations of a compound, generating a dose-response curve. I can calculate mean and sd of the replicates till the cows come home, but now I need to show a statistical difference between two concentrations on the curve. Statistics class was decades ago, and my book assumes I’ve read it from page 1 when I get to the t-test part (p. 315). I’m using Excel to calculate the t-test, and my questions are:

1. should I use a one- ot two-tailed test? I was thinking of a two-tailed test for greater stringency.

2. I assume I should do a paired test, as opposed to a homo- or heteroscedastic test, right? I am comparing two sets of the same number of data points from the three experiments done in triplicate (so N = 9).

3. how do I determine what p value to indicate? p\<0.01, 0.05, 0.001? I was never clear as to what degree of separation was indicated by the notation.

Thanks in advance.

Vlad +/- Igor, p\<certain.

---

<div class="post-metadata">

**Author:** ![Vlad\_Igor](https://avatars.discourse-cdn.com/v4/letter/v/7feea3/32.png) [@Vlad\_Igor](https://boards.straightdope.com/u/Vlad_Igor)\
**Post date:** [December 10, 2007, 2:28pm UTC](https://boards.straightdope.com/t/need-help-with-statistics/429289/2 "2007-12-10T14:28:05Z")

</div>

Noodge. 57 views and no one can help me? There’s gotta be someone out there who knows how to calculate p values.

---

<div class="post-metadata">

**Author:** ![BetsQ](https://avatars.discourse-cdn.com/v4/letter/b/fbc32d/32.png) [@BetsQ](https://boards.straightdope.com/u/BetsQ)\
**Post date:** [December 10, 2007, 3:10pm UTC](https://boards.straightdope.com/t/need-help-with-statistics/429289/3 "2007-12-10T15:10:59Z")

</div>

Um, I’m not exactly a statistics expert, but my gut responses to your questions are:

1. Yes, two-tailed. (In the social sciences that’s the default unless you have some compelling reason for using a one-tailed test.)

2. From your description of the project, I don’t think this sounds like a paired sample. You need to have pairs of individual data points, not just two groups of data.

3. Can’t you use a statistical program that will give you p-values? Stata or SPSS or something? I don’t have much SPSS experience, but I know you can just import a csv file and it’s not difficult to run t-tests.