Detecting multidimensionality in Rasch data using Winsteps Table 23

Подписаться 1,1 тыс.

Просмотров 10 тыс.

50% 1

Principal Components Analysis of Residuals (PCAR) is used to discover secondary dimensions in the unexplained variance after the Rasch dimension has been removed from ordinal data. Guidance is given as to when secondary dimensions are problematic and what action can be taken.
Missing data? Missing data are assigned residuals of zero when the correlations are calculated.
This video follows the method of www.rasch.org/rmt/rmt191h.htm but applies it to your measure distributions instead of to hypothetical measure distributions, so your simulation is more exact for your situation.
www.winsteps.com

Наука

Опубликовано:

6 авг 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 29

@scientific-accomplishment 6 месяцев назад

I am always respecting you!!

@shahdue75 3 года назад

Thank you very much for this tutorial. I've watched it a number of times. When I re-watch (to analyse a different set of data), I gained a better understanding.

@PittuS79 5 лет назад

Thank you Dr. Linacre. This is of great help and great update on analyzing dimensionality

@fouzulkareemaismail1835 3 года назад

Thank you, Dr. Linacre. This video is very helpful.

@jreub84 Год назад

This is incredibly helpful! Thank you for sharing!

@joaquincruz2404 2 года назад

Lovely! Finally, I got to understand it!!

@googie200011 5 лет назад

This is very helpful, thank you. g

@frohnzy04 Год назад

brilliant Mike. Thank you.

@johnmichaellinacre4960 Год назад

You asked: hi mike, can you tell me how to make the graph for info in table 23.6 and plot the trend line. It's not from laziness I ask, but speed. I am still coming to grips with this info and PCA and developing winsteps and now excel skills. A beautiful...but long journey alone. TIA Reply: copy and paste Table 23.6 into Excel. It will all go in column A. Then in Excel, Data, Text to Columns, Fixed Width. Then with the columns of numbers you want. Excel , Insert, Scatterplot, and add a Trend line to the scatterplot. OK?

@johnmichaellinacre4960 Год назад

You say you do not have Table 23.1 - strange! With Table 23.6, you can compute the disattenuated correlations yourself - see www.rasch.org/rmt/rmt101g.htm

@frohnzy04 Год назад

Hi Mike, you are incredible generous with your time and help. Thank you for these replies. Re. Tale 23.1. I was using a version on winsteps from a training course. I just bought the new version. I have that table now.

@frohnzy04 Год назад

I will follow your advice on using 23.6. thank you again.

@tomgallacher6968 6 лет назад

John - very useful. To clear up a confusion: person measures from items of Cluster 1 versus person measures from items of Cluster 3 correlated at -.17, and in the simulated the same correlation was .2, that's quite difficult to compare! so do we conclude that the -.17 was expected given the data since it's nearer zero, or do we conclude that -.17 was so far from the .2 that the -.17 was not expected given the data. If we're simulating according to a unidimensional Rasch model would we expect all person measures from item clusters to be positively correlated? To me the negative correlation is enough to conclude weirdness without checking against the equivalent correlation from simulated data. e.g. parallel analysis would compare eigenvalues against those expected given simulated data, but eigenvalues can't go negative.

@johnmichaellinacre4960 6 лет назад

Thanks, Tom. We definitely expect the person measures produced by different clusters of items from roughly the same construct to be positively correlated. The null hypothesis would not be a correlation of zero, but a disattenuated correlation of 1.0. However, we have deliberately stratified the items to capitalize on differences, so the null hypothesis is rather less than 1.0, which is also what the simulated data tell us. And, yes, these data are deliberately weird, so we should have noticed problems earlier in the analysis, hopefully long before Table 23!

@PittuS79 5 лет назад

Hello and thank you for the helpful video. When I run Rasch analysis I do not get the correlation table (with pearson and dissattenuated correlation values) following table 1. Can you please suggest how to reproduce those outputs. Also how can i reporduce the excel graph. I went through the manual and I do not find the suggested on my winsteps screen option through output tables or plots ? Please guide

@johnmichaellinacre4960 5 лет назад

Hi Priti, please check that you are using a recent version of Winsteps. The correlations are in Table 23.0 in Winsteps 3.75 (2012) onwards.

@kimlombard1638 4 года назад

Hi Mike, this was a great video, thank. My version of WINSTEPS is 3.68.2, is it not possible to do several of these steps in this version? For example, there is no Table 23.1. Thanks in advance, Kim

@johnmichaellinacre4960 4 года назад

Hi Kim, yes. Winsteps 3.68 and other earlier versions of Winsteps have the Table 23 subtables numbered differently. For instance, current Table 23.1 is probably Table 23.2 in your version. Current Table 23 includes information not in earlier Table 23. See www.winsteps.com/winman/ for information about extra information in Table 23.

@kimlombard1638 4 года назад

@@johnmichaellinacre4960 Hi Mike, Thank you very much for your informative response, this is good to know, thanks! Kim

@kimlombard1638 4 года назад

@@johnmichaellinacre4960 Hi Mike, one last question, if I was to purchase the newest edition of WINSTEPS, is it advisable that I uninstall my current version before installing the new version? Many thanks, Kim

@johnmichaellinacre4960 4 года назад

@@kimlombard1638 Hi Kim, no need to uninstall :-)

@learn5081 5 лет назад

Can we do this in FACETS as well?

@johnmichaellinacre4960 5 лет назад

Sorry, not in Facets, but Facets can output a Winsteps Control and Data file for PCAR analysis. See Facets Help - www.winsteps.com/facetman/winstepsscreen.htm

@zinibrahim1026 4 года назад

In table 23.2, the red box looks like the Knox Cube data while the blue box looks like the Liking for Science. Did you purposely combine the two sets of data to show there are 2 dimensions?

@johnmichaellinacre4960 4 года назад

Yes, well noticed!

@ignatiuslien2715 3 года назад

Dr Linacre, i tried to import in an excel file but got an error,; WININPUT 4.6.0 at excelmod line: 22070 Error: 1AD ActiveX component can't create object. Could i check why is this so? I ensure there is only 1 tab in the excel file

@johnmichaellinacre4960 3 года назад

Hi Ignatius: thanks for trying, and for asking :-) Winsteps needs a fully-installed Windows version of Excel. Even then sometimes a Microsoft component is missing, So also please install www.winsteps.com/a/Winsteps-Facets-system-files.exe - download this, then right-click on it and "Run as Administrator". OK, Ignatius?

@ignatiuslien2715 3 года назад

Thank you Dr Linacre, I was wondering if winstep could be install on Mac, i tried it without success.

@johnmichaellinacre4960 3 года назад

@@ignatiuslien2715 Sorry, Winsteps is 32-bit windows software. It does not work, even under Windows, in recent iOS versions.