In SPSS Factor analysis one component has both - and + values. I found that component values can be interpreted as regular correlation. So I don't get how variables in the same factor can have different directions of correlation. What can I do with those factors?
Hi Rūta,
It is much commun we have factors with factorial load + and -. This fact can be attribit two sitaations:
1. You have variables with ideas in negative direction. For exemple:
(a) "We can play the fry oil in domestic sewer"
versus
(b) "We cannot play the fry oil in domestic sewer"
If you have a construct for measurement the enviorement conscience, case (a) will give correlation (load factorial) negative and in case (b) positive
In those cases you should change the variable with negative ideia, inverting the sequence of measurement ( 5 - total desagree to 1 total agree - for exemple)
2. really, the variable with negative load is express a opositive idea, when we compareted this variable with other variables. In this case, do the interpretation. The respondents desagreed.
In two cases look at the mean value for confirmated which variable is really negative or has opositive propose
Best regards
Dirceu
Ruta,
For clarifications, the factor analysis is not the same as the principal component analysis. If you are trying to measure an invisible construct, you want to focus on the common variance of the items. In SPSS, the principal axis factoring will provide an option to analyze the common variance. The principal component analysis includes all variances including the common, unique and error variances. This issue has been pointed out repeated for a long time but hardly fixed. If you want some references, please check on Russell (2002) or Reise et al. (2000).
Regarding your question, Dirceu already pointed out the correct direction. Before you run the factor analysis, you should recode reverse coded items so that all correlation coefficients are in the same direction.
Hopefully, this helps!
Hello, thank you both for suggestions. It definitelly helps somewhat :)
Professor Da Silva, I see that I do not have situation number 1: my data already is all sigle direction elements with Likert scale. Situation number 2 i do not understand. How come elements do come to the same group if they are evaluated opositelly. Shouldn't they be similarly evaluated to be together (or something similar)?
Professor Moon, if I have all variables evaluated with Likert scale (no missing, nothing special, plain regular all cases 1 to 5), in such case then is Principal components method ok to use?
Thank you so much for your time. Please have patience with me if my questions seem too basic. I haven't studied a course on Factoral analysis, just some basic descriptives and Pearson on SPSS almost ten years ago :)
It is likely that you are measuring an invisible construct if you are using a Likert type scale but it is hard to tell until I learn what you intend to measure. The characteristics of the data are not the critical issue here. In most social science studies, PCA (principal component analysis) can be rarely justified. So what are you trying to measure?
Hi Rūta,
In a real research about the food marketing we had the follow factor loads (Likert scale with five points to choice):
Assertive (Factor load)
2. 1. I like a well-groomed table. (-.658)
2. 2. Even when I have company, I am very practical because I do not worry about the way I serve and to eat. (.757)
2. 4. Eating is just like any other task of the day (.750)
2.5. I do not like to lose time on meals (.621)
At first moment we can believ that assertive "I like a well-groomed table" showed a desagree tendence by the respondents (factor load is negative ==> -.658)
But when we analyse the means arrived in inverse results. Look:
Assertive (Factor load) [Mean]
2. 1. I like a well-groomed table. (-.658) [4.47]
2. 2. Even when I have company, I am very practical because I do not worry about the way I serve and to eat. (.757) [2.41]
2. 4. Eating is just like any other task of the day (.750) [2.34]
2.5. I do not like to lose time on meals. (.621) [2.19]
thus, the interpretation is inverse. the respondents "like a well-groomed table" (mean > 2.5) and disagree of three others (means
Dear Professor da Silva,
thank you so much for your explanaition in depth. Unfortunatelly due to my weak background in statistics this raises more questions than gives answers.
I solved my "problem" by using fixed number of factors. With 10 factors I don't have any factors including only two elements and no strong "-" values.
Dear Professor Moon,
what you are saying worrys me a bit about my findings. As I have found rather satisfying outcomes with Principal component analysis (the main one for Factor analysis in textbooks I have and that is the reason I chose it), you persisting on Principal axis factoring makes me wonder. With Principal axis factoring I get completely different outcome. Outcome which I cannot call meaningfull results.
I have found the Factor analysis helps to determine whicht elements (statements) are evaluated mostly similar. That is what i did and with Principal component analysis I have 10 groups of elements where each group has a clear common ground.
Could you please be more specific on WHY Principal component analysis in social sciences is not ok?
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS