Can I use time and gender as covariates in Ancova?
Locked
 
|
Hi all, I am studying social implications of media use from 2008 to 2010, three time waves. Time 1, time 2 and time 3
I want to check the mean differences on several independent variables, and belive Ancova is a could way. Is it possible to use gender and age as covriates in Ancova? Another thing is, can I use time 1 as co-variates in the the second wave analysis? How do I do this? Anyone? Thanks in advance. best petter |
Re: Can I use time and gender as covariates in Ancova?
|
Do the respondents participate in each of the three waves? Or are there
3 non-overlapping groups of respondents? or what? gender would usually be treated as a factor (independent variable) rather than as a covariate. How do you measure age? What are your other independent? what are their levels of measurement? What is/are your dependent variable(s)? How measured? How many respondents do you have? Art Kendall Social Research Consultants On 12/15/2010 10:52 AM, PetterBB wrote: > Hi all, I am studying social implications of media use from 2008 to 2010, > three time waves. Time 1, time 2 and time 3 > > I want to check the mean differences on several independent variables, and > belive Ancova is a could way. Is it possible to use gender and age as > covriates in Ancova? > > Another thing is, can I use time 1 as co-variates in the the second wave > analysis? How do I do this? Anyone? > > Thanks in advance. > > best > petter > > > > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Can-I-use-time-and-gender-as-covariates-in-Ancova-tp3306510p3306510.html > Sent from the SPSSX Discussion mailing list archive at Nabble.com. > > ===================== > 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
Art Kendall
Social Research Consultants |
|
|
Do the respondents participate in each of the three waves? Or are there
3 non-overlapping groups of respondents? or what? The groups are overlappin, the same users participate in all three waves (N = 391) gender would usually be treated as a factor (independent variable) rather than as a covariate. I know, but would rather see if 5 different groups of social networking sites user, sporadics, lurker, socializers, debaters and advanced differ in their social capital., using age and gender as covariates How do you measure age? 5 different categoriers from 15 to 75 years What are your other independent? what are their levels of measurement? The indpendent are the 5 clusters of users (found by cluster analysis) sporadics, lurker, socializers, debaters and advanced What is/are your dependent variable(s)? How measured? I am using different measures on social captial. 4 and 5-point Likert scales are used. The different items in each measure are added togheter. How many respondents do you have? N=391, all particpate in all 3 time waves. |
Re: Can I use time and gender as covariates in Ancova?
|
|
What I understand so far.
You have as predictor/right side/independent variables TYPE with 5 levels. Strictly nominal. Between respondents. AGE with 5 levels. At least ordinal. Between respondents Gender with 2 levels. Nominal. Between respondents. TIME with 3 levels. At least ordinal. WITHIN respondents(i. e., repeated) . predicted/dependent variable(s). SocCap a social capital summative score(s) across a series of items. That sounds like a 5 TYPE by 5 AGES by 2 Gender by 3 times ANOVA. A 5*5*2 *3R design Why would you not want to control for some of the effect of age and gender? It seems that is should be an empirical question whether interactions effects are needed in the model. If you crosstab TYPE by AGE by Gender do you have any empty cells? Art Kendall Social Research Consultants On 12/15/2010 12:44 PM, PetterBB wrote: > Do the respondents participate in each of the three waves? Or are there > 3 non-overlapping groups of respondents? or what? > > The groups are overlappin, the same users participate in all three waves (N > = 391) > > gender would usually be treated as a factor (independent variable) > rather than as a covariate. > > I know, but would rather see if 5 different groups of social networking > sites user, sporadics, lurker, socializers, debaters and advanced differ in > their social capital., using age and gender as covariates > > How do you measure age? > 5 different categoriers from 15 to 75 years > > What are your other independent? what are their levels of measurement? > The indpendent are the 5 clusters of users (found by cluster analysis) > sporadics, lurker, socializers, debaters and advanced > > What is/are your dependent variable(s)? How measured? > I am using different measures on social captial. 4 and 5-point Likert scales > are used. The different items in each measure are added togheter. > > How many respondents do you have? > N=391, all particpate in all 3 time waves. > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Can-I-use-time-and-gender-as-covariates-in-Ancova-tp3306510p3306675.html > Sent from the SPSSX Discussion mailing list archive at Nabble.com. > > ===================== > 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
Art Kendall
Social Research Consultants |
|
|
In reply to this post by PetterBB
You have as predictor/right side/independent variables =>Yes TYPE with 5 levels. Strictly nominal. Between respondents. => Yes AGE with 5 levels. At least ordinal. Between respondents => Yes, but I have also a age variable with 15 levels, So I have two age variables. Gender with 2 levels. Nominal. Between respondents. =>Yes TIME with 3 levels. At least ordinal. WITHIN respondents(i. e., repeated) => Yes predicted/dependent variable(s). SocCap a social capital summative score(s) across a series of items. => Yes That sounds like a 5 TYPE by 5 AGES by 2 Gender by 3 times ANOVA. A 5*5*2 *3R design => Yes Why would you not want to control for some of the effect of age and gender? => Yes, I will, but I thought the covariates could manage this. How can I do this otherwise. It seems that is should be an empirical question whether interactions effects are needed in the model. => Gender and age is of importance If you crosstab TYPE by AGE by Gender do you have any empty cells? => No I don't get empty cells. Art Kendall Social Research Consultants |
|
|
Hi I thought that with the use of ANCOVA the analysis will adjust the means of the outcome variable to what they would be if all subjects scored equally on the covariates; gender and age....it this wrong?
best Petter Bae Brandtzæg. |
Re: Can I use time and gender as covariates in Ancova?
Administrator
|
One of the problems you're running into in this thread is terminological. In SPSS lingo, covariate = continuous (or scaled) explanatory variable, and factor = categorical explanatory variable. So the variables you are describing as "covariates" are being seen by Art (and others) as "factors", because you have age categories, not age as a continuous variable. Do you have the actual ages? If so, why are you treating it as categorical?
--
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/). |
|
|
Thanks for your comment. Unfortunately, I do not have the actual age, only age described in 15 different categories.....will this then mean that I can not use this as covariates? Do you have any advice?
best petter
2010/12/16 Bruce Weaver [via SPSSX Discussion] <[hidden email]>
-- Petter Bae Brandtzæg Ustvedtsvei 14b O871 Oslo Tlf + 4792806546 |
Re: Can I use time and gender as covariates in Ancova?
|
Administrator
|
You can include them as "fixed factors", as Art suggested on Dec 15.
--
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/). |
Re: Can I use time and gender as covariates in Ancova?
|
|
In the ANCOVA only the main effects of age and gender would be "controlled"/"accounted" for. The ANOVA would additionally control for the interaction effects of (age by gender), � (age by time), (gender by time), and (age by gender by time). If you only did the ANCOVA part of the overall ANOVA, the 4 effects above would be pooled in the residual term (i.e., error) making it larger.� The ANCOVA is less powerful that the full ANOVA. Think of a test statistic as effect divided by error so the larger the error the smaller the test statistic. Art Kendall Social Research Consultants On 12/16/2010 9:22 PM, Bruce Weaver wrote: ===================== 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 REFCARDYou can include them as "fixed factors", as Art suggested on Dec 15. PetterBB wrote:Thanks for your comment. Unfortunately, I do not have the actual age, only age described in 15 different categories.....will this then mean that I can not use this as covariates? Do you have any advice? best petter 2010/12/16 Bruce Weaver [via SPSSX Discussion] < [hidden email][hidden email]PetterBB wrote: Hi I thought that with the use of ANCOVA the analysis will adjust the means of the outcome variable to what they would be if all subjects scored equally on the covariates; gender and age....it this wrong? best Petter Bae Brandtzæg. One of the problems you're running into in this thread is terminological. In SPSS lingo, covariate = continuous (or scaled) explanatory variable, and factor = categorical explanatory variable. So the variables you are describing as "covariates" are being seen by Art (and others) as "factors", because you have age categories, not age as a continuous variable. Do you have the actual ages? If so, why are you treating it as categorical? -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." *NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. * ------------------------------ View message @ http://spssx-discussion.1045642.n5.nabble.com/Can-I-use-time-and-gender-as-covariates-in-Ancova-tp3306510p3308697.html To unsubscribe from Can I use time and gender as covariates in Ancova?, click here<-- Petter Bae Brandtzæg Ustvedtsvei 14b O871 Oslo Tlf + 4792806546----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Can-I-use-time-and-gender-as-covariates-in-Ancova-tp3306510p3308881.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 PetterBB
Petter,
In addition to the advice you have received thus far, let me suggest you consider employing a linear MIXED model instead of an general linear model. I am making this suggestion since each participant was measured on the outcome of interest three times. It isn't uncommon to observe residuals obtained from observations closer in time to be more highly correlated than residuals obtained from more distant observations. Thus, an autoregressive residual correlation structure may be in order. In addition to decaying residual correlations, it's possible to have subject specific intercepts. Both specifications can be made via the MIXED procedure in SPSS. Whether these specifications substantially improve model fit can be examined. Ryan On Wed, Dec 15, 2010 at 10:52 AM, PetterBB <[hidden email]> wrote: > Hi all, I am studying social implications of media use from 2008 to 2010, > three time waves. Time 1, time 2 and time 3 > > I want to check the mean differences on several independent variables, and > belive Ancova is a could way. Is it possible to use gender and age as > covriates in Ancova? > > Another thing is, can I use time 1 as co-variates in the the second wave > analysis? How do I do this? Anyone? > > Thanks in advance. > > best > petter > > > > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Can-I-use-time-and-gender-as-covariates-in-Ancova-tp3306510p3306510.html > Sent from the SPSSX Discussion mailing list archive at Nabble.com. > > ===================== > 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 Art Kendall
Art Kendall, thanks, but I am not sure if I understand you.
Do you mean it is wrong to use ANCOVA in with my dataset? My main goal is to check wether or not different user styles in social networking sites imply different types of social capital, by using gender and age as covariates. However, if I do the ANOVA the way you suggest, is this done by 2way ANOVA or is it 1way? I am not sure how I can control for interaction effects in ANOVA the way you are suggesting it? Do you have any good instructions in how I can do this? Thanks. Petter |
Re: Can I use time and gender as covariates in Ancova?
|
|
I am suggesting that you "control for" all of the sources of variance
that you have measures for. That is a 4 way ANOVA. Type * Age * Gender * time. It is inefficient not to control for all of the known sources of variance. Those sources that are not in the design are therefore in the residual (error). All of the effects in the ANCOVA are in the ANOVA. The ANOVA is more complete. The ANCOVA controls for only some of the effects of age and gender. The interaction effects are taken care of by the software when you specify the design. Below there is a simulation of such an analysis, Open a new instance of SPSS. copy and paste the syntax. Run it. You can generate the syntax via the GUI. Art Kendall Social Research Consultants * make up some data. new file. set seed 20101219. input program. loop id = 1 to 391. compute pre = rv.normal(1,1). compute mid = pre + rv.normal(1,1). compute post = mid + rv.normal(1,1). end case. end loop. end file. end input program. compute type =1+ mod(id,5). compute gender =1+ mod(id,2). compute age = rnd(rv.uniform(1,5)). formats id (f3) type gender age (f1). value labels type 1 'a' 2 'b' 3 'c' 4 'd' 5 'e'/ gender 1 'male' 2 'female'/ age 1 'first' 2 'second' 3 'third' 4 'fourth' 5 'fifth'. *check that there are no empty cells. crosstabs tables = type by gender by age. * -----------------. *show the kind of analysis. * Custom Tables. CTABLES /VLABELS VARIABLES=type gender age pre mid post DISPLAY=LABEL /TABLE type > gender > age + gender BY pre [MEAN] + mid [MEAN] + post [MEAN] /CATEGORIES VARIABLES=type gender age ORDER=A KEY=VALUE EMPTY=INCLUDE. GLM pre mid post BY type gender age /WSFACTOR=time 3 Polynomial /CONTRAST(age)=Polynomial /METHOD=SSTYPE(3) /PLOT=PROFILE(time time*type time*gender time*age time*age*gender) /EMMEANS=TABLES(OVERALL) /EMMEANS=TABLES(type) COMPARE ADJ(BONFERRONI) /EMMEANS=TABLES(gender) COMPARE ADJ(BONFERRONI) /EMMEANS=TABLES(age) COMPARE ADJ(BONFERRONI) /EMMEANS=TABLES(time) COMPARE ADJ(BONFERRONI) /EMMEANS=TABLES(type*gender) /EMMEANS=TABLES(type*age) /EMMEANS=TABLES(type*time) /EMMEANS=TABLES(gender*age) /EMMEANS=TABLES(gender*time) /EMMEANS=TABLES(age*time) /EMMEANS=TABLES(type*gender*age) /EMMEANS=TABLES(type*gender*time) /EMMEANS=TABLES(type*age*time) /EMMEANS=TABLES(gender*age*time) /EMMEANS=TABLES(type*gender*age*time) /PRINT=ETASQ OPOWER /CRITERIA=ALPHA(.05) /WSDESIGN=time /DESIGN=type gender age type*gender type*age gender*age type*gender*age. On 12/18/2010 10:45 AM, PetterBB wrote: > Art Kendall, thanks, but I am not sure if I understand you. > > Do you mean it is wrong to use ANCOVA in with my dataset? My main goal is to > check wether or not different user styles in social networking sites imply > different types of social capital, by using gender and age as covariates. > > However, if I do the ANOVA the way you suggest, is this done by 2way ANOVA > or is it 1way? I am not sure how I can control for interaction effects in > ANOVA the way you are suggesting it? Do you have any good instructions in > how I can do this? > > Thanks. > Petter > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Can-I-use-time-and-gender-as-covariates-in-Ancova-tp3306510p3310522.html > Sent from the SPSSX Discussion mailing list archive at Nabble.com. > > ===================== > 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
Art Kendall
Social Research Consultants |
|
|
Thanks, this was helpful, do I need to re-code the age-variable or can I put all the variables in the analysis as it is?
petter
2010/12/19 Art Kendall [via SPSSX Discussion] <[hidden email]>
-- Petter Bae Brandtzæg Ustvedtsvei 14b O871 Oslo Tlf + 4792806546 |
Re: Can I use time and gender as covariates in Ancova?
|
|
In the example syntax I posted note that 5 levels were created for age.� Also that polynomial contrasts were specified because age is at least ordinal. Art Kendall Social Research Consultants On 12/19/2010 5:01 PM, PetterBB wrote: Thanks, this was helpful, do I need to re-code the age-variable or can I put all the variables in the analysis as it is?===================== 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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |