SPSS and Questionnaire
Locked
 
|
Hi everybody,
I have a question about my data analysis in SPSS and I was wondering if you could possibly help me out. I have a validated questionnaire with 19 items with a Likert scale of 1 to 6. Hence, the total score obtained from this questionnaire would be 19 to 114. The first question is about how to divide the questionnaire into "low, average, and high? I think running an ANOVA and selecting one SD below and above the mean would help to choose the average and the score below and above that as showing low and high. Is that right? I need to add that the questionnaire is a "language identity measure" and my research question for this part is: What is the degree of English language learners’ language identity measured through the language identity questionnaire? Is that low, moderate, or high? What do you think would be the best way to choose the mid point and also what would be the best way to answer this question. For me I received the mean from the SPSS output showing something like 54.7 which could possibly fall in the average side. What do you think?Thanks in advance for your kind attention. I will be looking forward to hearing from you. -- Regards Saeed Rezaei Lecturer Languages and Linguistics Center Sharif University of Technology Tehran IRAN Skype: [hidden email] |
Automatic reply: SPSS and Questionnaire
|
|
I am in an training class Wednesday and Thursday, 5-6 December. I will respond to emails in the evening. Please contact:
·
[hidden email] for questions about or assistance with MIP ProjectPages
(page functionality, content on the pages, i.e. PI name, proejct title, etc.)
·
[hidden email] for questions about charge
numbers
·
[hidden email]
for quesitons or assiatcne with MIP processes or the webiste
·
[hidden email] for assistance with
other Innovation Zone sites, such as CI&T InZone Regards, Mary Lou |
|
In reply to this post by Saeed Rezaei
1: Best to subtract 19 from your total score to yield a variable with a true zero point. File > New > syntax [to open a syntax file] then copy/modify the syntax examples below] Your syntax needs to look like this (substitute your variable names): Compute score = sum.19 (<v1> to <v19>) – 19. This only creates a score for people who have answered all 19 items: all other total scores will be missing. 2: Forget Anova: just run a frequency count on the values of score and check the cumulative % column. You can choose your cutting points empirically or you can ask for percentiles. Freq score /per 33 67 /sta mea med /his nor. . . will give you cutting points to divide into three approx equal groups, mean median and a histogram overlaid with a normal curve. Use RECODE to create your groups (substitute the <values> you choose) Recode score (0 thru <value1> = 1) (<value1 thru <value2> = 2) (<value2> thru 114 = 3) into group. Value labels group 1 ‘Low’ 2 ‘Medium’ ‘High’ 3. Freq group /bar. I’m shortly off to UK for 10 days so won’t be here to follow this up. Have a look at the SPSS tutorials on my site for further enlightenment. John F Hall (Mr) Email: [hidden email] Website: www.surveyresearch.weebly.com From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Saeed Rezaei Hi everybody, What is the degree of English language learners’ language identity measured through the language identity questionnaire? Is that low, moderate, or high? What do you think would be the best way to choose the mid point and also what would be the best way to answer this question. For me I received the mean from the SPSS output showing something like 54.7 which could possibly fall in the average side. What do you think? |
|
In reply to this post by Saeed Rezaei
What are you going to
do with the transformed variable? Are you
thinking of relating it to some other variable(s)?
Depending on why items are missing you might want to get an average score. If there is a reason to leave it on the 19 to 114 scale you could "blow up" the mean. something like this UNTESTED syntax should do it. compute score1 = mean.12(item1 to item19). compute score2 = rnd( mean.12(item1 to item19) * 19). compute score3 = sum.19(item1 to item19. variable labels score1 'mean item score (on the original 1 to 6 response scale)' score2 'blown up mean item score ( 19 to 114)' score3 'summative score for people with all 19 items (19 to 114)'. *check shape of score distributions. *replace "respondentID" with the variable that identifies your case. explore variables = score1 t score3 /compare = variables /id = respondentID /percentiles = (.333333, .666667) /plot= boxplot histogram npplot /statistics = descriptives extreme (10). *transform scores into 3 ntiles. rank variables = score1 to score3 /ntiles(3) into thirds1 to thirds3. value labels third1 to thirds3 1 'low' ... Art Kendall Social Research ConsultantsOn 12/5/2012 5:14 AM, Saeed Rezaei wrote: Hi everybody, ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Art Kendall
Social Research Consultants |
Administrator
|
In reply to this post by Saeed Rezaei
Carving scaled variables into categories is usually a bad idea. See David Streiner's nice article, for example:
http://www.ncbi.nlm.nih.gov/pubmed/11987478 https://ww1.cpa-apc.org/publications/archives/cjp/2002/april/streiner.PDF Why wouldn't you just treat it as a continuous variable? HTH.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
In reply to this post by Saeed Rezaei
Saeed,
What I'm about to propose would require an entire rethinking of how to construct a scoring system for this type of questionnaire. Assuming you are, in fact, measuring a unidimensional construct [which can be empirically evaluated<--not a question but a statement], a good place to start would be to determine exactly where the items and individuals fall along the continuum of the construct. Perhaps some items are more "difficult" to endorse than others. That is, individuals with the same true trait level of "language identity" may have an easier time endorsing some items as compared to other items. Maybe, just maybe, some items reflect higher degrees of "language identity" than others. Why then would we derive a summative score where items are weighted equally? ...And then compound the problem by trying to create cut-off scores based on a potentially flawed scoring system? Moreover, the assumption that response options are ordered may not be true. In general, total scores lack the type of sensitivity/accuracy for which contemporary test developers strive. Look up latent trait theory. Ryan On Wed, Dec 5, 2012 at 5:14 AM, Saeed Rezaei <[hidden email]> wrote: > > Hi everybody, > I have a question about my data analysis in SPSS and I was wondering if you could possibly help me out. > I have a validated questionnaire with 19 items with a Likert scale of 1 to 6. Hence, the total score obtained from this questionnaire would be 19 to 114. The first question is about how to divide the questionnaire into "low, average, and high? I think running an ANOVA and selecting one SD below and above the mean would help to choose the average and the score below and above that as showing low and high. Is that right? > I need to add that the questionnaire is a "language identity measure" and my research question for this part is: > > What is the degree of English language learners’ language identity measured through the language identity questionnaire? Is that low, moderate, or high? > > What do you think would be the best way to choose the mid point and also what would be the best way to answer this question. For me I received the mean from the SPSS output showing something like 54.7 which could possibly fall in the average side. What do you think? > > Thanks in advance for your kind attention. > > I will be looking forward to hearing from you. > > -- > Regards > Saeed Rezaei > Lecturer > Languages and Linguistics Center > Sharif University of Technology > Tehran > IRAN > Skype: [hidden email] > > > ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
Saeed,
I am not against determining *substantively meaningful* cut-off scores of a sensitive/accurate/reliable measure of a psychological construct, for clinical purposes. What I am against (as Bruce alluded to) is categorizing continuous variables for analysis without good reason. These are two completely separate issues. Ryan On Wed, Dec 5, 2012 at 7:47 AM, Saeed Rezaei <[hidden email]> wrote: > Thanks for your replies everybody. Do you think transforming "language > identity" into a "measure" a flaw? If so why do we have other fuzzy > constructs such as "anxiety" and "critical thinking" and....all translated > into quantifiable concepts? > > What about categorizing the construct into “low, medium, and high”? Is that > problematic too? What do you suggest as a substitute? > > > > > > On Wed, Dec 5, 2012 at 11:32 PM, R B <[hidden email]> wrote: >> >> Saeed, >> >> What I'm about to propose would require an entire rethinking of how to >> construct a scoring system for this type of questionnaire. Assuming >> you are, in fact, measuring a unidimensional construct [which can be >> empirically evaluated<--not a question but a statement], a good place >> to start would be to determine exactly where the items and individuals >> fall along the continuum of the construct. Perhaps some items are more >> "difficult" to endorse than others. That is, individuals with the same >> true trait level of "language identity" may have an easier time >> endorsing some items as compared to other items. Maybe, just maybe, >> some items reflect higher degrees of "language identity" than others. >> Why then would we derive a summative score where items are weighted >> equally? ...And then compound the problem by trying to create cut-off >> scores based on a potentially flawed scoring system? Moreover, the >> assumption that response options are ordered may not be true. In >> general, total scores lack the type of sensitivity/accuracy for which >> contemporary test developers strive. >> >> Look up latent trait theory. >> >> Ryan >> >> On Wed, Dec 5, 2012 at 5:14 AM, Saeed Rezaei <[hidden email]> wrote: >> > >> > Hi everybody, >> > I have a question about my data analysis in SPSS and I was wondering if >> > you could possibly help me out. >> > I have a validated questionnaire with 19 items with a Likert scale of 1 >> > to 6. Hence, the total score obtained from this questionnaire would be 19 to >> > 114. The first question is about how to divide the questionnaire into "low, >> > average, and high? I think running an ANOVA and selecting one SD below and >> > above the mean would help to choose the average and the score below and >> > above that as showing low and high. Is that right? >> > I need to add that the questionnaire is a "language identity measure" >> > and my research question for this part is: >> > >> > What is the degree of English language learners’ language identity >> > measured through the language identity questionnaire? Is that low, moderate, >> > or high? >> > >> > What do you think would be the best way to choose the mid point and also >> > what would be the best way to answer this question. For me I received the >> > mean from the SPSS output showing something like 54.7 which could possibly >> > fall in the average side. What do you think? >> > >> > Thanks in advance for your kind attention. >> > >> > I will be looking forward to hearing from you. >> > >> > -- >> > Regards >> > Saeed Rezaei >> > Lecturer >> > Languages and Linguistics Center >> > Sharif University of Technology >> > Tehran >> > IRAN >> > Skype: [hidden email] >> > >> > >> > >> >> ===================== >> To manage your subscription to SPSSX-L, send a message to >> [hidden email] (not to SPSSX-L), with no body text except the >> command. To leave the list, send the command >> SIGNOFF SPSSX-L >> For a list of commands to manage subscriptions, send the command >> INFO REFCARD > > ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by Saeed Rezaei
Be sure to consider
Bruce' response .
He was more explicit in pointing out that coarsening data is often counterproductive. Was the scale previously determined to be unidimensional? Am I correct that this is more an attitude scale than an ability scale? Art Kendall Social Research ConsultantsOn 12/5/2012 5:14 AM, Saeed Rezaei wrote: Hi everybody, ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Art Kendall
Social Research Consultants |
|
|
In reply to this post by Art Kendall
Perhaps you should tell
us more about your project.
e.g., Are you doing research on these questions and still figuring out what specific statistical tools to use? Or have you committed to others that you would use a coarsened measure and specific statistical tools? Or are you replicating previous research that used specific statistical tools? Or ... ---------------- the way your research question is stated I see no reason to coarsen the data before analysis. ----------- Even for very beginners the menu's etc are a great way to draft syntax, it is very productive to use the menu for a procedure until you like the way it works. Then go back to the menu for that procedure and without changing anything exit via <paste>. there are many reasons for using syntax. * doing an analysis is a process that goes through revisions and refinement. * As you progress through an analysis you will refine how you are thinking about the problem. You will develop your skills and you will develop your understanding of the particular data you are working with. * You will be interrupted during an analysis. Meals, bathroom breaks, phone calls, etc. etc. * syntax is necessary for supervision and for the due diligence of quality assurance review. * it clarifies for you what you are instructing the computer to do. * it provides a way for people who might replicate you research to know what you did. * it communicates to people whom you are asking for help exactly what the problem is. * by providing an audit trail syntax helps YOU understand what you did when you go back to the analysis. Art Kendall Social Research ConsultantsOn 12/5/2012 7:35 AM, Saeed Rezaei wrote: The transformed variable is intended to be used for other research questions i.e. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Art Kendall
Social Research Consultants |
|
|
In reply to this post by Saeed Rezaei
Please copy the
conversation to the SPSSx-l list. This enables
other members to add useful suggestions and considerations. It
also builds up the archive so that other people with questions can
see the whole conversation.
It is surprising that you are using very sophisticated techniques like AMOS but do not yet use syntax! Please explain why you want to use a coarsened version of your measure. What specific statistical tools are you considering? Art Kendall Social Research ConsultantsOn 12/5/2012 8:06 AM, Saeed Rezaei wrote: This is an "attitude scale" more with six sub-components as done through EFA and CFA through SEM (AMOS). The construct is "language identity" with six sub-components namely "attachment to your language, pronunciation attitude", language and social status, L1 use in the society, language knowledge, and alphabet". These components were verified through AMOS and the model fitness indices. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Art Kendall
Social Research Consultants |
|
|
In reply to this post by Art Kendall
Hi again
Thanks for all your comments. I went through a number of steps to develop a hypothesized model at first. The model was "language identity Model for English language learners in Iran" After reviewing the literature on the topic and referring to the relevant theories and consultations with the experts int eh field I hypothesized a model with "six main components as I told you before. Then to test the model I had to develop an instrument and since there were no viable instrument in the literature fulfilling my purpose I developed a questionnaire with 3-5 items tapping each of the subcomponents of the construct I had (language identity). After going through a dozend of changes in the questionnaire, it was administered for reliability estimate through Cronbach Alpha which exhibited a respectable degree of reliability for both the whole questionnaire and its sub-parts. After ensuring about the reliability, EFA and CFA was run on the questionnaire with the adminsitartion of the questionnaire in two separate sessions and the six factors were extracted but some minor changes and tailoring happened there.The initial AMOS fit indices showed that the model was not fit but a second time it was run with some changes in the model and finally it showed fairly good indices. Now that I have developed and validated the questionnaire, I need to administer it to the Iranian English language learners to answer my research questions. My research questions for this phase are: 1. What is the degree of Iranian English language learners’ language identity measured through the language identity questionnaire? Is that low, moderate, or high? 2. Is there any relationships between their language identity level and their demographic characteristics including their · language proficiency level · age, · and gender? Now...I really enjoyed your comments. What do you suggest that I should do to answer these research question?Thanks Saeed On Thu, Dec 6, 2012 at 12:31 AM, Art Kendall <[hidden email]> wrote:
|
|
|
I don't understand the construct you are attempting to measure, assuming it is a construct [which is still in question for me]. At any rate, I carefully thought out my response before sending it this morning. I've yet to receive any answers. If you want a way to identify cut-points that are meaningful along the continuum of the construct you are attempting to measure, then you will need to re-read my post. As a side note, if you choose to remain in the CTT realm, and you decide that your construct is multidimensional, then coefficient alpha will misestimate reliability of the composite (total) scores, given the assumptions upon which it is based. Ryan
|