Skip to main content

First post using stackedit

Welcome to StackEdit! Hey! I’m your first Markdown document in StackEdit1. Don’t delete me, I’m very helpful! I can be recovered anyway in the Utils tab of the Settings dialog.
Documents StackEdit stores your documents in your browser, which means all your documents are automatically saved locally and are accessible offline!
Note:
StackEdit is accessible offline after the application has been loaded for the first time.Your local documents are not shared between different browsers or computers.Clearing your browser’s data may delete all your local documents! Make sure your documents are synchronized with Google Drive or Dropbox (check out the Synchronization section). Create a document The document panel is accessible using the button in the navigation bar. You can create a new document by clicking New document in the document panel.
Switch to another document All your local documents are listed in the document panel. You can switch from one to another by clicking a document in the li…

Testing the difference between two correlations in R

Introduction


Correlation is the measurement to measure how two datasets increase or decrease together. It's a real number range from -1 (i.e, when one dataset increases, another one decreases and vice versa) to +1 (i.e, these two datasets increase or decrease together). Value 0 means there is no relationship between two datasets.

Calculating correlation


Calculating correlation of two datasets in R is quite straightforward:

In R:

> x = c (1,5,2,6,7)
> y = c (2,4,3,9,-5)
> cor(x,y)
[1] -0.1459338

Usually, a single value does not say much in statistics. You need a confidence level.

> cor.test (x, y)

Pearson's product-moment correlation

data:  x and y
t = -0.2555, df = 3, p-value = 0.8149
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.9109173  0.8451475
sample estimates:
       cor 
-0.1459338 

The function 'cor.test' will gives you the correlation, and also 95% confidence level of two datasets.

Testing two correlations

Now, we move to the next part: testing two correlations, to see whether they are different or not.

In this section, I will use the package "cocor". In order to use this package, you need to install it.

> install.packages ("cocor")
> library (cocor)

The following example comes from cocor documentation, which can be found at http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0121945

> data (“aptitude”)

> cocor (~logic + intelligence.a | logic + intelligence.a, aptitude)
 Results of a comparison of two correlations based on independent groups
Comparison between r1.jk (logic, intelligence.a) = 0.3213 and r2.hm (logic, intelligence.a) = 0.2024
Difference: r1.jk—r2.hm = 0.1189
Data: sample1: j = logic, k = intelligence.a; sample2: h = logic, m = intelligence.a
Group sizes: n1 = 291, n2 = 334
Null hypothesis: r1.jk is equal to r2.hm
Alternative hypothesis: r1.jk is not equal to r2.hm (two-sided)
Alpha: 0.05
fisher1925: Fisher’s z (1925)
z = 1.5869, p-value = 0.1125
Null hypothesis retained
zou2007: Zou’s (2007) confidence interval
95% confidence interval for r1.jk—r2.hm: -0.0281 0.2637

Null hypothesis retained (Interval includes 0)

Please do not forget the sign ~ at the beginning of the formula.
From the result, we can conclude whether two correlations are different or not (again, with a confidence interval)





Comments

Popular posts from this blog

Installing tensorflow on Mac OS 10.11

Tensorflow (https://www.tensorflow.org/) is the open source deep learning library from Google (http://www.wired.com/2015/11/google-open-sources-its-artificial-intelligence-engine/).

Installing Tensorflow is not easy on Mac if you follow exactly the installation instruction on the homepage. I have no idea why. I have error then error while trying to install Tensorflow with pip, with virtualenv, or even with Docker.

However, with Anaconda (https://www.continuum.io/downloads), it will be easy. The trick is you should download the wheel file and install it offline other than retrieving it online as suggested by Tensorflow homepage.

So, with anaconda for python2 installed, I did:

# create a new environment with sklearn installed, up to you
# if you want a pure Python, replace scikit-learn by python
conda create -n tensorflow scikit-learn

# activate the new environment
source activate tensorflow

# download the wheel file

curl https://storage.googleapis.com/tensorflow/mac/tensorflow-0.6.0-py2-none-…

How to use diffent keyboard than English in Ubuntu Server 14.04 (or 14.10) with Gnome Core

Even in my laptop, I still use Ubuntu Server because it seems more stable than Ubuntu Desktop.

At the first time of using Ubuntu Server, you may have some issues, especially if you are not familiar with Linux command, because there is no GUI in Ubuntu Server by default.

There are some ways to install GUI to Ubuntu Server.

I am choosing Gnome Core.

To install Gnome Core:

sudo apt-get install xorg gnome-core gnome-system-tools gnome-app-install
If you have used Ubuntu Desktop before (with Unity), you may find something different in Gnome Core.

1. The notification bar now in on the bottom but not top.
 To see the notification bar, move your mouse to the bottom of the screen.

2. There is no Text Entry anymore, so if you are not native English speaker like me and want to install another typing application such as ibus to type in your own language, it may be pain.

In Gnome Core, to configure your typing, press Windows button (if there is this key in your keyboard), or click to Activities at…