Attached 3-4 pages for your reading.

I think it's very helpful.

Note: However as Daniel suggested you must have different time points to help decide the function that relates effect size and time

Effect size is measure of the strength of a phenomenon.

for example- correlation between two variables, the regression coefficient in a regression, the mean difference.

The absolute effect size is the difference between the average, or mean, outcomes in two different intervention groups

A larger absolute value of effect size always indicates a stronger effect.

So, in point 1. we can't say both have same effect size.

I hope that following article be useful, plz check out.

Regards

Obviously, you can extrpolate effect size if you have strong theoretically and empirically backed assumtions about the development of effects. In your example, I would not expect a linear development, but something resembling the S-shaped learning curve (although this is just a guess). So, I would say two measurement points in time are not enough for an educated guess about the trend of effect size. If you have longitudinal data with 4 or 5 measurements, it should be different.

You need more time points to help decide the function that relates effect size and time. Linear is unlikely for most interventions in my field (but may be reasonable in your field).

Steven;

I am not sure which method you are using to calculate your effect size, but it appears that the sizes fall between small, medium, and large according to the attached article, so you can use this as a benchmark to determine whether your effect sizes are similar.

Best regards,

Debra

http://www.bwgriffin.com/gsu/courses/edur9131/content/Effect_Sizes_pdf5.pdf

Extrapolation is possible only you know the shape of the impact curve. In most intervention studies the curve is NOT linear so you cannot extrapolate that double time will be double impact. For example when taking anyibiotics the effect size with 2 weeks of medication is high but 4 weeks will not double. In my fields -education and psychology- the curve is almost never linear and you cannot extrapolate easily.

Thank you everyone so very much for your responses and the further readings.  Greatky appreciated.  Steve

You're welcome. I also learned more about effect size from responses posted, so thanks in return for asking the question.

Merry Christmas and Happy New Year to all!

Debra

Attached 3-4 pages for your reading.

I think it's very helpful.

Note: However as Daniel suggested you must have different time points to help decide the function that relates effect size and time

Good question Steve. Alart from the general statistical meaning and manipulation of effect size, we have been pondering the actual meaning of effect size taken under variable conditions as you describe. For example, in pre/post tests of limited scope over 5 weeks, i often get an effect size as large as 1. Since hattie described an effect size of 0.4 as 'normal', does that mean i am a brilliant teacher or is it a side effect of my methodology. If the variability in effect size is a consequence of methodology then its use has little meaning for the classroom teacher. Anther questions that arose was "Is effect size sensitive to the type of learning tested? eg a focus on low order content vs higher order thinking." If it is sensitive to this then it could 'nudge' teachers/systems towards favouring one over the other. My experiernce suggests that a focus on low order content yields higher effect size with worrying possible consequences.

I try to make the concept of effect size as simple as possible for students:

Think of the IQ scale with mean of 100 and standard deviation of 15.  For effect size we compare the mean of the experimental group to the mean of the control group.  The difference is divided by the standard deviation of the control group (whatever that empirical amount is).  The resulting number (effect size) is the proportion comparable to the proportion of the SD of the IQ scale.  For example, effect size d=3 is the equivalent of a 4.5-point rise in the IQ scale, d=5 is equivalent to 7.5 IQ scale points and d=.7 is the equivalent of 10.5 points on the IQ scale.  These values have equivalents on the percentile scale as well so a conversion chart is handy:  IQ 110 (d=.7) is at the 75th %ile (25-percentile point rise of experimental  compared to the control mean).

IQ score of 115 (d=1.0) is at the 84th percentile.  For the difference in means expressed in percentile equivalent, subtract 50 %ile points so that d=1.0 is an increase of 34 percentile points.  Many different scales exist, and the z-score scale is the scale for standard deviations and fractions thereof.  Learning to think in terms of equivalents in the different scales is fundamental. Most people understand IQ points and effect size (d=) is a z-score difference at the mean of 0 difference (equivalent of 50th %ile, IQ=100, T-score of 50, etc.) .  Visualizing the normal curve and the various scales that extend along the base of the curve is essential.  Effect is a small area extending from the center of the curve (the mean) to the right as explained above.

The importance of an effect varies with the phenomenon sampled. Effect usually can be discussed in terms of the study only.  Extrapolation is hypothetical and is without evidence (but providing a question to be explored).  Extrapolation is more of a question than a probable answer.

Note: In early childhood (kindergarten) children can make twice the progress rate in the second semester compared to the first semester, so extrapolation would be highly inaccurate with linear increments; final months in a school year can produce large increases since students have mastered learning tools that help increase the rate of learning.

As with all experiments at the beginning, take the time to think about and answer these questions:  What do you want to be able to say? What will it take to support that statement?  

The readings would suggest that time cannot be used to extrapolate. In fact effect size can only be compared when there is a significant degree of comparability between 2 interventions or between an intervention and control. So in the case I described myself above, my effect sizes are standalone. They mean that something has happened. It was positive, but i can't say with accuracy that it was large, medium or small. A confounding factor is the emotional impact on my interventions. The curriculum topic and structure, time of year, temperature of day etc all influenced the results, so comparing them was meaningless.

Let me be simple (and I am not a statistician).

As John Hattie argued, significant effect size in education starts at .40

Thus:

1. Regardless of the number of students, the difference between the effect size of the two studies is significant, thus, not at all 'similar".   Said that, I think it is a mistake to compare the two groups.

2. You might. But, I predict a disastrous extrapolation/prediction. People are not machine. Events are not lineal and they are more complex than a lineal cause-effect attribution

The effect size of d=.4 would be the equivalent of 6 points improvement of the mean on the IQ scale, since each decimal fraction equals 1.5 IQ scale points.  When influences can be combined, as in one intervention producing an increase of .2 and another intervention .3 both might be worth including.  

Note:  in small groups (classrooms, e.g.), the median is considered a better central tendency indicator than means, Informally 

Dear Steven

Cautiously, I think the discussion continues - and needs to continue - on this issue, so I think it unwise to give a "settled answer," so to speak. My own take is: Sometimes Yes; Sometimes No. Have a look at: 

How Science Takes Stock: The Story of Meta-Analysis: The Story of Meta-Analysis

Av Morton Hunt. 

Best wishes in this important research, Paul