Funded through the ESRC’s Researcher Development Initiative Session 3.3 & 3.4: Teacher Expectancy Example Prof. Herb Marsh Ms. Alison O’Mara Dr. Lars-Erik Malmberg Department of Education, University of Oxford Session 3.3 & 3.4: Teacher Expectancy Example Establish research question Define relevant studies Develop code materials Data entry and effect size calculation Pilot coding; coding Locate and collate studies Main analyses Supplementary analyses (Meta-analysis data from Raudenbush & Bryk, 2002) Do teacher expectations influence student IQs? Teachers led to have high expectations of experimental (through bogus feedback) but not control students. The focus is on the effect of how long teachers knew students prior to the experimental intervention. Teacher Expectancy Effects on IQ (Metaanalysis data from Radudenbush & Bryk, 2002 Do teacher expectations influence student IQs? Teachers led to have high expectations of experimental (through bogus feedback) but not control students. Focus here is on the effect of how long teachers knew students prior to the experimental intervention. Teacher Expectancy Effects on IQ (Metaanalysis data from Radudenbush & Bryk, 2002 6 Potential Outliers 90th %tile 75th %tile Median 25th %tile 10th %tile Mean effect size and homogeneity analyses SPSS Commands File: TCH-EXPT meta-analysis 19cases.sav compute w = 1/SE ** 2. Include "N:\MetaAnalysis\SPSS\METAES.SPS" MeanES ES=d /W=w. MeanES MACRO 2523 2524 2525 2526 2527 2529 2530 2532 2534 2535 2536 2537 2539 2540 2541 2542 2543 2545 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 *-------------------------------------------------------------*' Macro for SPSS/Win Version 6.1 or Higher *' Written by David B. Wilson (dwilson@crim.umd.edu) *' Meta-Analyzes Any Type of Effect Size *' To use, initialize macro with the include statement: *' INCLUDE "[drive][path]MeanES.SPS" . *' Syntax for macro: *' MeanES ES=varname /W=varname /PRINT=option . *' E.g., MeanES ES = D /W = IVWEIGHT . *' In this example, D is the name of the effect size variable *' and IVWEIGHT is the name of the inverse variance weight *' variable. Replace D and INVWEIGHT with the appropriate *' variable names for your data set. *' /PRINT has the options "EXP" and "IVZR". The former *' prints the exponent of the results (odds-ratios) and *' the latter prints the inverse Zr transform of the *' results. If the /PRINT statement is ommitted, the *' results are printed in their raw form. MeanES MACRO MeanES ES=d /W=w. ------- Distribution Description --------------------------------N Min ES Max ES Wghtd SD 19.000 -.320 1.180 .218 ------- Fixed & Random Effects Model ----------------------------Mean ES -95%CI +95%CI SE Z P Fixed .0603 -.0112 .1319 .0365 1.6539 .0981 Random .0893 -.0201 .1987 .0558 1.6005 .1095 ------- Random Effects Variance Component -----------------------v = .025920 ------- Homogeneity Analysis ------------------------------------Q df p 35.8254 18.0000 .0074 Random effects v estimated via noniterative method of moments. MeanES MACRO MeanES ES=d /W=w. Conclusions: Small (NS) effect size based on both Fixed & Random models. Significant unexplained variance suggesting non-generalisability of effects across studies and the need for a random effects model. ------- Distribution Description --------------------------------N Min ES Max ES Wghtd SD 19.000 -.320 1.180 .218 ------- Fixed & Random Effects Model ----------------------------Mean ES -95%CI +95%CI SE Z P Fixed .0603 -.0112 .1319 .0365 1.6539 .0981 Random .0893 -.0201 .1987 .0558 1.6005 .1095 ------- Random Effects Variance Component -----------------------v = .025920 ------- Homogeneity Analysis ------------------------------------Q df p 35.8254 18.0000 .0074 Random effects v estimated via noniterative method of moments. Analogue to the ANOVA analyses MetaF MACRO (ANOVA) based on categorical weeks METAF AF=d /W=w /group = wkcat/MODEL=ML . Conclusions: Large effect Q) ------of weeks (20.4/35.8= p .0001 57% var expl); .4198 .0074 Total Residual, variance component & residual by group all NS; ES significant for 1st two groups, NS last two groups ------- Analog ANOVA table (Homogeneity Q df Between 20.3788 3.0000 Within 15.4466 15.0000 Total 35.8254 18.0000 ------- Q by Group ------Group Qw df p .0000 6.2140 4.0000 .1837 1.0000 4.9825 2.0000 .0828 2.0000 .4045 2.0000 .8169 3.0000 3.8456 7.0000 .7974 ------- Effect Size Results Total ------Mean ES SE -95%CI +95%CI Z P k Total .0603 .0365 -.0112 .1319 1.6539 .0981 19.0000 ------- Effect Size Results by Group ------Group Mean ES SE -95%CI +95%CI Z P k .0000 .3620 .1099 .1466 .5774 3.2939 .0010 5.0000 1.0000 .3516 .1121 .1319 .5712 3.1367 .0017 3.0000 2.0000 .0665 .0727 -.0760 .2091 .9147 .3604 3.0000 3.0000 -.0630 .0500 -.1611 .0350 -1.2605 .2075 8.0000 ------- Maximum Likelihood Random Effects Variance Component ------v = .00000 se(v) = .00648 Regression analyses MetaReg MACRO (Regression) based of categorical weeks METAREG ES=d /IVS =/W=w wkcat /MODEL=ML . Conclusions: Large ------- Descriptives ------effect of weeks Mean ES R-Square k (54% var expl); .0603 .5375 19.0000 Constant term ------- Homogeneity Analysis ------highly significant Q df p (at intercept = 0); Model 19.2573 1.0000 .0000 Residual variance Residual 16.5680 17.0000 .4840 NS (variance Total 35.8254 18.0000 .0074 component = 0) ------- Regression Coefficients ------B SE -95% CI +95% CI Z P Beta Constant .4072 .0871 .2366 .5779 4.6775 .0000 .0000 -.1573 .0358 -.2275 -.0870 -4.3883 .0000 -.7332 ------- Maximum Likelihood Random Effects Variance Component ------v = .00000 se(v) = .00648 MetaReg MACRO (Regression) based on uncategorical weeks METAREG ES=d /W=w/IVS=weeks /MODEL=ML . Conclusions: Large effect of weeks (21% var expl); ------- Descriptives ------Mean ES R-Square k .0686 .2107 19.0000 Constant term highly ------- Homogeneity Analysis ------significant (at Q df p intercept = 0); Model 6.7540 1.0000 .0094 Residual var NS; Residual 25.3035 17.0000 .0881 Does not do as well as Total 32.0575 18.0000 .0216 categorised weeks ------- Regression Coefficients ------B SE -95% CI +95% CI Z P Beta Constant .1622 .0550 .0544 .2699 2.9498 .0032 .0000 weeks -.0132 .0051 -.0232 -.0032 -2.5988 .0094 -.4590 ------- Maximum Likelihood Random Effects Variance Component ------v = .00501 se(v) = .00895 Variations of the previous analyses MetaF MACRO (ANOVA) based of Blind vs. Aware Test Administrators METAF ES=d /W=w /group = BLDvAWR/MODEL=ML . ------- Analog ANOVA table (Homogeneity Q) ------Conclusions: Small, NS Q df p Between .8381 1.0000 .3599 effect; Within 26.8855 17.0000 .0598 Resid var marginally Total 27.7236 18.0000 .0664 significant for “Aware” ------- Q by Group ------not “blind” Group Qw df p 1.0000 9.9550 8.0000 .2682 2.0000 16.9305 9.0000 .0498 ------- Effect Size Results Total ------Mean ES SE -95%CI +95%CI Z P k Total .0785 .0480 -.0156 .1727 1.6344 .1022 19.0000 ------- Effect Size Results by Group ------Group Mean ES SE -95%CI +95%CI Z P k 1.0000 .1314 .0752 -.0159 .2788 1.7487 .0803 9.0000 2.0000 .0420 .0625 -.0805 .1644 .6720 .5016 10.0000 ------- Maximum Likelihood Random Effects Variance Component ------v = .01333 se(v) = .01263 MetaF MACRO (ANOVA) based of Group vs. Individual IQ Tests METAF ES=d /W=w /group = GPvIN /MODEL=ML . Conclusions: Small, marginally significant ------- Analog ANOVA table (Homogeneity Q) ------Q df p effect (4.1/34.3=12% Between 4.1421 1.0000 .0418 var expl); ES NS for Within 30.1044 17.0000 .0256 “Group” but signi for Total 34.2464 18.0000 .0117 “individual” ------- Q by Group ------Resid var for group Group Qw df p signif but individual NS; 1.0000 25.2512 15.0000 .0467 2.0000 4.8531 2.0000 .0883 ------- Effect Size Results Total ------- All effects very small Mean ES SE -95%CI +95%CI Z P k Total .0638 .0386 -.0118 .1393 1.6536 .0982 19.0000 ------- Effect Size Results by Group ------Group Mean ES SE -95%CI +95%CI Z P k 1.0000 .0444 .0397 -.0335 .1222 1.1169 .2640 16.0000 2.0000 .3812 .1607 .0663 .6960 2.3726 .0177 3.0000 ------- Maximum Likelihood Random Effects Variance Component ------v = .00189 se(v) = .00745 MetaReg MACRO (Regression) based of Group vs. Individual IQ Tests METAREG ES=d /W=w/IVS=GPvIN /MODEL=ML . Conclusions: Small, marginally significant effect (12% var expl); ------- Descriptives ------Mean ES R-Square k .0638 .1209 19.0000 Constant term ------- Homogeneity Analysis ------Q df p (Group) NS; Model 4.1421 1.0000 .0418 Resid var signif (but Residual 30.1044 17.0000 .0256 variance component Total 34.2464 18.0000 .0117 NS) ------- Regression Coefficients ------B SE -95% CI +95% CI Z P Beta Constant -.2924 .1792 -.6437 .0588 -1.6318 .1027 .0000 GPvIN .3368 .1655 .0124 .6612 2.0352 .0418 .3478 ------- Maximum Likelihood Random Effects Variance Component ------v = .00189 se(v) = .00745 MetaReg MACRO (Regression) based of (Group vs. Individual IQ Tests) & (categorised weeks) Conclusions: Effect of METAREG ES=d /W=w/IVS=GPvIN wkcat /MODEL=ML . test type no longer ------- Descriptives ------signif when weeks Mean ES R-Square k included. .0603 .5770 19.0000 ------- Homogeneity Analysis ------Effect of weeks nearly Q df p unaffected. Model 20.6711 2.0000 .0000 Note can only look at Residual 15.1543 16.0000 .5134 multiple variables Total 35.8254 18.0000 .0074 with regression. ------- Regression Coefficients ------B SE -95% CI +95% CI Z P Beta Constant .1782 .2113 -.2360 .5925 .8434 .3990 .0000 GPvIN .1983 .1667 -.1286 .5251 1.1890 .2344 .2032 wkcat -.1481 .0367 -.2199 -.0762 -4.0399 .0001 -.6904 ------- Maximum Likelihood Random Effects Variance Component ------v = .00000 se(v) = .00648 Website Address to get MLwiN Harvey Goldstein developed the MLwiN statistical package used here and has made many contributions to multilevel modeling, including meta-analysis. Always a bit dangerous to say some one person invented a new approach. However, fair to say that Stephen Raudenbush at least popularised the multilevel approach to meta-analysis with the meta-analysis of the teacherexpectancy data considered here. Raudenbush, S.W. and Bryk, A.S. (2002).Hierarchical Linear Models (Second Edition).Thousand Oaks: Sage Publications, 482 pp. Raudenbush, S.W. (1984). Magnitude of teacher expectancy effects on Pupil IQ as a function of the credibility of expectancy induction: A synthesis of findings from 18 experiments.Journal of Educational Psychology, 76, 1, 85-97. Getting Data Into MLwiN In an empty MLwiN file, puts the xx input variables into first xx columns. (can also add new data to existing files). Check to see that data is correct and click on “paste” button Getting Data Into MLwiN Check the MLwiN “names” file to see that data looks ok (e.g., missing values; min & max values). Setting Up Meta-analysis MLwiN will open an empty equation that you have to construct. Click on the “y” to bring up this screen. select “d” (the effect size) as the dependent variable Select “2” for “N of levels” select “ID” for Level 1 select “d” for Level 2 Setting Up Meta-analysis 3 2 1.Click “Add Term” Button (bottom equations window) 2.Select “cons” (variable = 1 for all cases) 3.Click the “done” button 1 Setting Up Meta-analysis 1 1.Click “Cons” in the 2 equation 2.Tick “Fixed Parameter” & “j(id)” but not “i(d)” 3 3.Click the “done” button Setting Up Meta-analysis 2 3 1 4 1.Now click “add term” button 2. This will bring up the “X-Variable” select SE (the standard error computed earlier) 3.Tick only the “i(d)” box 4. Click “done” Setting Up Meta-analysis 1 1 Now we want to constrain the variance at level 1 to be fixed at 1.0. Under “model” select “constrain parameters”; will bring up “parameter constraint” window Setting Up Meta-analysis 1 In the parameter constraint window: 1.Click the “random” button 2.Change “d: SE/SE” to 1 3.Change “to equal” to 1” 2&3 Setting Up Meta-analysis 3 1 2 1. “store” the constraints in the first empty column (“C19”) 2. Click the “attach random constraints” button. 3. Close the “Parameter Constraint” Window “null” model with no predictors ->pred c50 ->calc c51=(('d'-c50)/'se')**2 ->sum c51 to b1 = 36.057 ->cprob b1 18 = 0.0069377 Conclusion: The mean effect size (.078) is not significant. The chi-square is significant; there is study-to-study variation. Reasonable to explore moderator variables After Closing the “parameter constraint” window (last slide) Click on “start” button in “equation” window (may have to click estimates button to get values). Compute chi-square value in command interface window Add raw “weeks” variable >pred c50 ->calc c51 = (('d'-c50)/'se')**2 ->sum c51 to b1 28.937 (Compared to 35.825 for null model) ->cprob b1 17 0.035115 Conclusion: The effect of weeks (-.013/.005) is significant The mean effect size (.162/.055) signif (when weeks = 0). chi-sq signif; some remaining study-to-study variation. WKCAT: 4-category weeks ->pred c50; ->calc c51 = (('d'-c50)/'se')**2;->sum c51 to b1 = 16.568 ->cprob b1 17 = 0.48398 Conclusion: categorized weeks does best of of (chi-sq = 16.568) Aware vs. Blind Administration For a categorical variable, you choose a reference (“left out” category. Default is the 1st category >pred c50; ->calc c51 = (('d' - c50)/'se')**2; ->sum c51 b1 = 35.608 ->cprob b1 17 = 0.0051714 Conclusion: Main Effect of Aware vs. Blind is NS Aware vs. Blind Administration For a categorical variable, you choose a reference (“left out” category. Default is the 1st category >pred c50; ->calc c51 = (('d' c50)/'se')**2; ->sum c51 b1 = 16.445 ->cprob b1 17 = 0.49254 Conclusion: Effect for Blind vs. Aware reduced by controlling for “wkcat” but was already nonsignificant Individual vs. Group Tests >pred c50; ->calc c51 = (('d'-c50)/'se')**2;->sum c51 to b1 = 31.478 ->cprob b1 16 = 0.011685 ->pred c50; ->calc c51 = (('d' - c50)/'se')**2; ->sum c51 b1 = 15.154 ->cprob b1 16 = 0.51337 Conclusion: Effect seems larger for individually administered tests, but not after control for weeks (wkcat) Individual vs. Group Tests Order = 1 to specify a 2-way interaction term Variables in interaction ->pred c50; ->calc c51 = (('d' - c50)/'se')**2; ->sum c51 b1 = 15.114 Conclusion: No interaction effect (chi-sq little different than wkcat alone (15.114 vs. 16.568). Notice that the effect of test type (and its SE) are very large (.2901/.4859=.597) Individual vs. Group Tests Grand Mean Centered Grand Mean Centered ->pred c50; ->calc c51 = (('d' - c50)/'se')**2; ->sum c51 b1 = 15.114 Conclusion: Same results but estimated & SE for individual term is smaller (reduced multicollinearity by grand mean centering the wkcat variable). Note that chi-sq is the same. “weeks” centered at 2 Weeks is centered at 2. >pred c50;->calc c51 = (('d'-c50)/'se')**2; ->sum c51 to b1 = 28.937 ; ->cprob b1 17 = 0.035115 Conclusion: The effect of weeks (-.013/.005) & chi-sq (28.937) same as with original weeks. The mean effect size (.136/.049) signif (when weeks = 2). “weeks” centered at 6 & 7 Weeks centered at 6 Weeks centered at 7 >pred c50;->calc c51 = (('d'-c50)/'se')**2; ->sum c51 to b1 = 28.937 ; ->cprob b1 17 = 0.035115 Conclusion: Constant term (intercept weeks = 6) is signif (.083/.042=1.98) Constant term (intercept weeks = 7) is NS (.070/.042=1.68) “weeks” polynomial = 2 >pred c50;->calc c51 = (('d'-c50)/'se')**2; ->sum c51 to b1 = 26.237; ->cprob b1 16 = 0.050779 Conclusion: The linear term is significant but the quad term is not. “weeks” polynomial = 3 ->calc c51 = (('d' - c50)/'se')**2; ->sum c51 b1 = 18.543; ->cprob b1 15 = 0.23521 Conclusion: All three polynomial terms are significant and the residual variance component is substantially reduced. “Log-e weeks+1” ->pred c50; ->calc c51 = (('d' - c50)/'se')**2; ->sum c51 b1 = 24.636 ->cprob b1 17 = 0.10317 Conclusion: The linear term based on the log transform explains more variance than the original (untransformed) weeks (chi-sq = 24.636 vs. 28.937). WKCAT: Centered at 2 & 3+ Conclusion: Intercept at wkcat = 2 is significant, but intercept at wkcat = 3 is not Graphs: Caterpillar Plots 1 Caterpillar plot based on L1 residuals. Go to the “model” menu and select “residuals” option. This will bring up the “settings” window. Set “SD (comparative)” to 1.96; 3. Set “level” to “1d”; 4. click the “Calc” button; 5. click on the “plot” button to bring up the next window. In the “plot” window select “residual +/- 1.96SD x rank. This brings up the original graph. Clicking on the graph bring up a window to modify the graph (a bit) The mean effect size associated with the intervention was not significant. However, the results did not generalise across studies (there was study-to-study variation). Consistent with a priori predictions, the effect size was significantly moderated by the amount of time students had been in contact with teachers. If students and teachers knew each other 0 or 1 week prior to the intervention, there was a significant expectancy effect. If they knew each other 2 or 3+ weeks the effect was not significant (although the precise cutoff might dependend on the scaling of weeks). Effects of test type (individual or group) and test administrator awareness (“blind” or aware) were not significant and did not interact with length. Purpose-built Comprehensive Meta-analysis (commercial) Schwarzer (free, http://userpage.fuberlin.de/~health/meta_e.htm) Extensions to standard statistics packages SPSS, Stata and SAS macros, downloadable from http://mason.gmu.edu/~dwilsonb/ma.html Stata add-ons, downloadable from http://www.stata.com/support/faqs/stat/meta.html HLM – V-known routine MLwiN MPlus Cooper, H., & Hedges, L. V. (Eds.) (1994). The handbook of research synthesis (pp. 521–529). New York: Russell Sage Foundation. Hox, J. (2003). Applied multilevel analysis. Amsterdam: TT Publishers. Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta-analysis: Correcting error and bias in research findings. Newbury Park: Sage Publications. Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. Thousand Oaks, CA: Sage Publications. Raudenbush, S.W. (1984). Magnitude of teacher expectancy effects on Pupil IQ as a function of the credibility of expectancy induction: A synthesis of findings from 18 experiments. Journal of Educational Psychology, 76, 85-97. Raudenbush, S.W. and Bryk, A.S. (2002). nd Hierarchical Linear Models (2 Ed.).Thousand Oaks: Sage Publications. Download macros for free from http://mason.gmu.edu/~dwilsonb/ma.html Download MLwiN for free from http://www.cmm.bristol.ac.uk/MLwiN/index.shtml