Subtracting scores on first principal component
|
|
Thread rating:  |
Duke - 01 Jul 2009 18:02 GMT I have done a PCA on 22 variables from a survey. All variables are measured on a scale of 1=strongly disagree to 7=strongly agree.
The first component contains positive loadings in the range 0.40 to 0.66 on all variables, and thus can be interpreted as a measure of size - some respondents tend to give higher scores in general than others. It accounts for 29% of the variation.
In this context, 'size' is a noise dimension, and the next 4 components, which account for far lower proportions of the variance (9% to 5%), appear to have an interpretable structure.
My question is: Is it an acceptable practice to 'standardise' the raw variables by subtracting the score on the first principal component per respondent? And if so, then to further use these 'standardised' variables as input into other multivariate techniques? And is there any literature to support this approach?
Your help would be greatly appreciated. Steve
Rich Ulrich - 01 Jul 2009 22:51 GMT >I have done a PCA on 22 variables from a survey. All variables are >measured on a scale of 1=strongly disagree to 7=strongly agree. [quoted text clipped - 13 lines] >variables as input into other multivariate techniques? And is there >any literature to support this approach? This is a little bit unintelligible. You are subtracting the score on the first component from WHAT?
Your next 4 components are uncorrelated with the first; "size" has been subtracted out of each of them.
If you want to control for "size" in further analyses with variables that are correlated with size, the usual step would be to include size as another covariate.
I wonder about this being a "survey" where size is so crucial. I have seen data, such as surveys of cities, where size (population) was a major correlate of things like "miles of streets", "number of police", "number of burglaries", etc. The conventional and proper pre-processing was one that converted the responses into units-per-capita or units-per-square-mile; or whatever. Not always easy, but appropriate.
 Signature Rich Ulrich
Ray Koopman - 01 Jul 2009 23:02 GMT > I have done a PCA on 22 variables from a survey. All variables are > measured on a scale of 1=strongly disagree to 7=strongly agree. [quoted text clipped - 16 lines] > Your help would be greatly appreciated. > Steve It's certainly not a standard practice, and probably not a good idea. The fact that all the variables load positively on the first (or any) component does not by itself imply that that component represents some sort of content-independent response bias such as 'willingness to agree'. There can be other explanations. This is not an appropriate forum for deciding the question. You need to talk to experts in the substantive area you are investigating.
Whatever the component may represent, if you do decide to remove it from the data then you don't simply subtract the component scores from the data. You partial the component scores out of the data. Treat the component as a predictor; regress each item in turn on the component, and keep the residuals. (There are various algebraic shortcuts, but I fear that giving them here would only confuse things.) The resulting scores will not be independent from subject to subject, and will to that extent be unsuitable for further standard analyses, but if the sample size is large then you may be able to ignore the problem and proceed as usual without worrying about it.
Duke - 02 Jul 2009 09:51 GMT Rich, Ray, thanks for your excellent responses.
Rich, to answer your question, what I was suggesting was to subtract the score on the first principal component per respondent from each of the 22 raw variable scores for that respondent. This of course only being thinkable because all of the original variables are measured on the same scale. Since the loadings are in a narrow range around 0.5, and the first PC is just another linear combination of the raw variables, this would probably give similar results to subtracting the mean score across all of the 22 raw variables per respondent from each of the 22 raw scores for that respondent. However, the particular linear combination chosen is data-driven and hence this may not be statistically reliable.
Rich, I take your point, the 'size' component should not be so dominant and we will have to make a call on whether or not there is too much noise in the data for it to be useable.
Ray, your suggestion of partialling the component scores out of the data is very helpful.
Regards, Steve
Duke - 02 Jul 2009 14:53 GMT PS obviously I meant with the first PC appropriately scaled so the sum of the weights = 1, as with the mean. Anyway, thanks for the help. Steve
Duke - 02 Jul 2009 15:01 GMT PS obviously I meant with the first PC appropriately scaled so the sum of the weights = 1, as with the mean. Anyway, thanks for the help. Steve
Art Kendall - 02 Jul 2009 22:33 GMT Are you sure you are looking at the rotated loadings?
Art Kendall Social Research Consultants
> I have done a PCA on 22 variables from a survey. All variables are > measured on a scale of 1=strongly disagree to 7=strongly agree. [quoted text clipped - 16 lines] > Your help would be greatly appreciated. > Steve Duke - 03 Jul 2009 09:17 GMT No, the unrotated principal component loadings. Steve
Art Kendall - 03 Jul 2009 13:44 GMT How many attitude constructs were the items designed to measure?
Are the items from an instrument used in earlier research?
If you are dealing with attitudes why did you choose a PC type of factor analysis?
What do the results look like when you use a more conventional approach? i.e., varimax rotation?
How do your eigenvalues compare to those from a parallel analysis? I.e., from doing a large number of runs (I tend to do 10,000) on random data with the same number of variables and cases? See the SPSSX-l archives for "parallel analysis" syntax.
Art Kendall Social Research Consultants
> No, the unrotated principal component loadings. > Steve iman islam - 04 Jul 2009 11:11 GMT What are the ‘Five Pillars’ of Islam? August 31st, 2008 I would be thankful if you give my article 5 minute of your value time.
THANK YOU
What are the ‘Five Pillars’ of Islam?
They are the framework of the Muslim life: faith, prayer, concern for the needy, self-purification, and the pilgrimage to Makkah for those who are able.
First Pillar: Faith
There is no god worthy of worship except God and Muhammad is His messenger. This declaration of faith is called the Shahada, a simple formula which all the faithful pronounce. In Arabic, the first part is la ilaha illa’Llah - ‘there is no god except God’; ilaha (god) can refer to anything which we may be tempted to put in place of God– wealth, power, and the like. Then comes illa’Llah: ‘except God’, the source of all Creation. The second part of the Shahada is Muhammadun rasulu’Llah: ‘Muhammad is the messenger of God.’ A message of guidance has come through a man like ourselves.
Second Pillar: Prayer
Salat is the name for the obligatory prayers which are performed five times a day, and are a direct link between the worshiper and God. There is no hierarchical authority in Islam, and no priests, so the prayers are led by a learned person who knows the Quran, chosen by the congregation. These five prayers contain verses from the Quran, and are said in Arabic, the language of the Revelation, but personal supplication can be offered in one’s own language.
Prayers are said at dawn, noon, mid-afternoon, sunset and nightfall, and thus determine the rhythm of the entire day. Although it is preferable to worship together in a mosque, a Muslim may pray almost anywhere, such as in fields, offices, factories and universities. Visitors to the Muslim world are struck by the centrality of prayers in daily life.
A translation of the Call to Prayer is:
‘God is most great. God is most great. God is most great. God is most great. I testify that there is no god except God. I testify that there is no god except God. I testify that Muhammad is the messenger of God. I testify that Muhammad is the messenger of God. Come to prayer! Come to prayer! Come to success (in this life and the Hereafter)! Come to success! God is most great. God is most great. There is no god except God.’
Once Muslims prayed towards Jerusalem, but during the Prophet’s lifetime it was changed to Makkah. From the minbar, the pulpit, the Imam who leads the prayer gives the sermon at the Friday noon community prayers.
Third Pillar: Zakat
One of the most important principles of Islam is that all things belong to God, and that wealth is therefore held by human beings in trust. The word zakat means both ‘purification’ and ‘growth’. Our possessions are purified by setting aside a proportion for those in need, and, like the pruning of plants, this cutting back balances and encourages new growth.
Each Muslim calculates his or her own zakat individually. For most purposes this involves the payment each year of two and a half percent of one’s capital. A pious person may also give as much as he or she pleases as sadaqa, and does so preferably in secret. Although this word can be translated as ‘voluntary charity’ it has a wider meaning. The Prophet (SAW) said: ‘Even meeting your brother with a cheerful face is charity.’
The Prophet (SAW) said: ‘Charity is a necessity for every Muslim.’ He was asked: ‘What if a person has nothing?’ The Prophet (SAW) replied: ‘He should work with his own hands for his benefit and then give something out of such earnings in charity.’ The Companions asked: ‘What if he is not able to work?’ The Prophet (SAW) said: ‘He should help poor and needy persons.’ The Companions further asked ‘What if he cannot do even that?’ The Prophet (SAW) said ‘He should urge others to do good.’ The Companions said ‘What if he lacks that also?’ The Prophet (SAW) said ‘He should check himself from doing evil. That is also charity.’
Fourth Pillar: The Fast
Every year in the month of Ramadan, all Muslims fast from first light until sundown, abstaining from food, drink, and sexual relations. Those who are sick, elderly, or on a journey, and women who are pregnant or nursing are permitted to break the fast and make up an equal number of days later in the year. If they are physically unable to do this, they must feed a needy person for every day missed. Children begin to fast (and to observe the prayer) from puberty, although many start earlier.
Although the fast is most beneficial to the health, it is regarded principally as a method of self purification. By cutting oneself off from worldly comforts, even for a short time, a fasting person gains true sympathy with those who go hungry as well as growth in one’s spiritual life.
Fifth Pillar: The Pilgrimage (Hajj)
The annual pilgrimage to Makkah, the Hajj, is an obligation only for those who are physically and financially able to perform it. Nevertheless, about two million people go to Makkah each year from every comer of the globe providing a unique opportunity for those of different nations to meet one another. Although Makkah is always filled with visitors, the annual Hajj begins in the twelfth month of the Islamic year (which is lunar, not solar, so that Hajj and Ramadan fall sometimes in summer, sometimes in winter). Pilgrims wear special clothes: simple garments which strip away distinctions of class and culture, so that all stand equal before God.
The rites of the Hajj, which are of Abrahamic origin, include circling the Ka’ba seven times, and going seven times between the mountains of Safa and Marwa as did Hagar during her search for water. Then the pilgrims stand together on the wide plain of Arafa and join in prayers for God’s forgiveness, in what is often thought of as a preview of the Last Judgment.
In previous centuries the Hajj was an arduous undertaking. Today, however, Saudi Arabia provides millions of people with water, modem transport, and the most up-to-date health facilities.
The close of the Hajj is marked by a festival, the Eid al-Adha, which is celebrated with prayers and the exchange of gifts in Muslim communities everywhere. This, and the Eid al-Fitr, a feast-day commemorating the end of Ramadan, are the main festivals of the Muslim calendar.
—————————————-
For more information about Islam
http://english.islamway.com/
http://www.islamhouse.com/
http://www.discoverislam.com/
http://www.islambasics.com/index.php
http://english.islamway.com/
http://www.islamtoday.net/english/
http://www.islamweb.net/ver2/MainPage/indexe.php
http://www.sultan.org/
Contact Us At
Imanway.group@gmail.com
cat.. - 05 Jul 2009 10:50 GMT > I have done a PCA on 22 variables from a survey. All variables are > measured on a scale of 1=strongly disagree to 7=strongly agree. [quoted text clipped - 16 lines] > Your help would be greatly appreciated. > Steve Hi Steve,
What I'm a bit concerned about in your approach is that you say that your data come from a survey. A PCA is based on the matrix of correlations. And correlations are to be weighted taking into account the sampling design. Did you take this into account in your code ?
Catherine.
|
|
|