HLM software has been one of the leading statistical packages for hierarchical


 Ernest Pitts
 6 years ago
 Views:
Transcription
1 Introductory Guide to HLM With HLM 7 Software 3 G. David Garson HLM software has been one of the leading statistical packages for hierarchical linear modeling due to the pioneering work of Stephen Raudenbush and Anthony Bryk, who created the software and authored the leading text on hierarchical linear and nonlinear modeling (Bryk & Raudenbush, 1992; Raudenbush & Bryk, 2002). Though differences among software packages capabilities have diminished over time, HLM 7 offers a number of appealing advantages and capabilities. Among these are what many consider to be a more intuitive model specification environment, greater ease in creating three and fourlevel models, its wide choice of estimation options, integrated likelihood ratio hypothesis testing, graphics options, and the ability easily to handle heterogeneous hierarchical linear models (where the dependent is thought to have different error variances for different levels of some grouping variable such as Agency). Scientific Software International (SSI) distributes HLM 7. A free student edition of HLM 7 is available. 1 The student edition is fullfeatured, including examples, but is limited in the size and complexity of models (though it will work with all example files provided with the software). HLM 7 software operates through several modules, each designed for a different type of HLM model, only some of which can be illustrated here due to space constraints: HLM2. For twolevel linear and nonlinear models with one dependent variable. HLM3 and HLM4. For threelevel and fourlevel models with one dependent variable. HGLM. For generalized linear models for distributions other than normal and link functions other than identity, handling binary, count, multinomial, and ordinal outcome variables in Bernoulli, binomial, Poisson, multinomial, and ordinal models. HLM SOFTWARE 55
2 56 PART I. GUIDE HMLM. For multivariate normal models with more than one outcome variable, including when the level 1 covariance structure is homogenous, heterogeneous, loglinear, or AR(1) (firstorder autoregressive). HMLM2. For twolevel HMLM models where level 1 is nested within level 2. HCM2. For models where level 1 units are crossclassified by two level 2 units. HCM3. For threelevel crossclassified models. HLMHCM. For two and threelevel hierarchical linear models with crossclassified random effects (ex., repeated test scores nested within students who are crossclassified by schools and neighborhoods). In summary, HLM 7 is a versatile and fullfeatured environment for many linear and generalized linear mixed models. ENTERING DATA INTO HLM 7 HLM software stores data in its own multivariate data matrix (MDM) format, which may be created from raw data or from data files imported from SPSS, SAS, Stata, SYSTAT, or other packages. MDM format files come in flavors keyed to the several types of HLM modules noted above. File creation options are accessed from the HLM File menu, illustrated in Figure 3.1 below. The example below illustrates data entry from an SPSS.sav file for models of type HLM2, but similar procedures are followed for other model types. Stat package input, depicted above, is the most common method of creating.mdm data files. Further, not only are data commonly prepared using statistical or data packages outside HLM 7, but as an additional preprocessing step, the researcher also should rule out multicollinearity among the level 2 (or higher) predictors. Having done this, there are two methods of importing files into HLM 7 from other statistical packages. Input Method 1: Separate Files for Each Level This method results in faster processing but requires more time to set up the data. It requires that separate files be created outside of HLM 7 for each level of HLM analysis. For SPSS, these are.sav files. For SAS, these are SAS 5 transport files. Separate SYSTAT and Stata files are also acceptable. For instance, HLM 7 software comes with example files from the Singer (1998) High School and Beyond study. The SPSS files for this example include HSB1.SAV, which contains the level 2 link field (ID is school ID) and any studentlevel variables. There are multiple rows per school, one row per student. It is critical that the level 1 file is sorted such that all students for a given school ID are adjacent.
3 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 57 Figure 3.1 HLM 7 file menu Likewise, the schoollevel (level 2) file, HSB2.SAV, contains the same level 2 link field and any schoollevel variables. Input Method 2: Using a Single Statistics Program Data File This method 2 is easier in terms of data management and is the one illustrated in this chapter. The same statistics package file formats as for Method 1 may be used. For the example, the single data file must be sorted such that all students for a given school ID are adjacent. Making the MDM File The next step is to create the.mdm file, which is HLM software s native data format. After it is created, the input data files are not needed. After creating the
4 58 PART I. GUIDE Figure 3.2 HLM 7 select MDM type window input data file in SPSS, SAS, or another package, HLM 7 is run and Stat package input is selected. This causes the Select MDM type window illustrated above to appear. The researcher chooses the HLM model type wanted. For instance, for a simple twolevel hierarchical linear model, the selection would be HLM2. After selecting HLM2, the Make MDM  HLM2 dialog box appears, illustrated in Figure 3.3. Here, the following steps are necessary: Set the Input File Type to SPSS/Windows (or another statistical package format). In the level 1 specification area, click the Browse button and browse to the input file for level 1. Then, as illustrated at the top of Figure 3.4, click the Choose variables button, click the checkbox indicating the level 2 link variable (id in the example), and click the checkboxes of any other level 1 variables in the analysis. In the level 2 specification area, click the Browse button and browse to the input file for level 2. This may be the same file as for level 1 (following Method 2 above). Again click the Choose variables button, click the checkbox indicating the level 2 link variable (agency), and click the checkboxes of any other level 2 variables in the analysis, as indicated in the lower half of Figure 3.4.
5 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 59 Figure 3.3 Make MDM  HLM 2 window in HLM 7 Save the MDM template file by clicking the Save mdmt file button, making sure the file location window points to the desired folder and giving a filename (add the.mdmt extension), then clicking the Save button. To complete the process, the researcher clicks the Make MDM button, giving a filename (here, M3_L2.mdm, standing for mixed linear model Chapter 3, 2level). The.mdm file is created, and the descriptive statistics module runs. Alternatively, one may click the Check Stats button. This output, shown in Figure 3.5, should be examined to verify the results. For instance, it is prudent to examine the reported sample size, which, if low, flags that the researcher has not sorted the Level 1 file to assure that individual rows for the same level 2 ID (AGENCY in this example) are adjacent.
6 60 PART I. GUIDE Figure 3.4 Choose variables windows in HLM 7
7 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 61 Figure 3.5 Check Stats button output in HLM 7 Click the Done button to exit to the WHLM model construction screen discussed below. At this point, the researcher will have saved three files to the disk: the newly created HLMcompatible data file, H3_L2.mdm in this example; the default template creatmdm.mdmt (the researcher may override the default name); and the output file above, HLM2MDM.STS (if desired, use File, Save As, to save output under a different name, as this default file may get reused with new content if there are multiple runs). After data are entered, the next step is to create the model. Typically, the first model created is the null model. The null model serves two purposes: (1) It is the basis for calculating the intraclass correlation coefficient (ICC), which is the usual test of whether multilevel modeling is needed; and (2) it outputs the deviance statistic (2LL) and other coefficients used as a baseline for comparing later, more complex models. For the current example, the null model addresses the question, Is there a (level 2) agency effect on the (level 1) intercept of performance score, which represents the mean score? If there is an agency effect, then ordinary regression methods will suffer from correlated error, and some form of linear mixed modeling is required. The null model, like all twolevel hierarchical models in HLM 7, is created in the WHLM modeling dialog, illustrated in Figure 3.6. This dialog is reached either on clicking Done in the MAKE MDM dialog or, if the MDM file was THE NULL MODEL IN HLM 7
8 62 PART I. GUIDE Figure 3.6 WHLM modeling window in HLM 7: Null model previously saved, from the HLM menu by selecting File, Create a new model using an existing MDM file, and then opening the appropriate.mdm file. In the WHLM modeling dialog illustrated in Figure 3.6, the employee level (level 1) dependent variable performance score (SCORE0) is designated as the outcome variable. No other predictors are added. HLM 7 already knows Agency is the level 2 grouping variable and automatically assumes Table 3.1 Summary of the Null Model it is a predictor of the level 1 intercept of SCORE0. When SCORE0 is designated as the outcome variable, HLM 7 constructs and displays the model, in this Level 1 Model SCORE0 ij = β 0j + r ij Level 2 Model β 0j = γ 00 + u 0j Mixed Model case the null model (also called the interceptonly model or the oneway ANOVA model with random effects). The null model is shown in Table 3.1 below. Clicking the Mixed button at the bottom of the WHLM dialog creates the combined HLM equation shown at the bottom of the figure: The two separate equations shown in the upper main window are mathematically equivalent SCORE0 to the single combined mixed model equation. For ij = γ 00 + u 0j + r ij learning purposes, it is easier to examine the equations
9 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 63 at each level. At level 1, SCORE0 is predicted by an intercept term and a random term. The symbol for the intercept term varies depending on the distribution specified for the outcome variable (this is done in the Basic Settings window, described below) and is expressed equivalently but differently in output. The level 1 intercept term, expressed as β 0j in output, is a function of a random intercept term at level 2 (γ 00 ) and a level 1 residual error term (r ij ). The level 1 intercept, in turn, is a function of the grand mean (γ 00 ) across level 2 units, which are agencies in this example, plus a random error term (u 0j ), signifying the intercept is modeled as a random effect. Substituting the righthand side of the level 2 equation into the level 1 equation gives the mixed model equation for the null random intercept model. HLM 7 will create one level 1 regression for each agency, and then will utilize the variance in these intercepts when estimating parameters and standard errors at level 1. This is what makes the process different from ordinary regression, where a single overall intercept is estimated. Before calculating estimates, the researcher may specify the distribution of the outcome variable by selecting Basic Settings from the main menu bar, yielding the window shown in Figure 3.7. The normal distribution, used in this example, is the default. Other available specifications support Bernoulli, Poisson, multinomial, and ordinal distributions. Selecting Bernoulli for a binary outcome variable applies a logistic link function, and in the ensuing multilevel logistic regression, interpretations are in terms of the log odds of the outcome rather than in terms of the raw outcome itself. Selecting Multinomial creates a multilevel multinomial regression using a logit link. Ordinal supports multilevel ordinal regression models. Multilevel Poisson regression models employ a Poisson log link and require an exposure variable (time, for example). In this window, one may also specify the name and location of the output statistics file and the output graphics file. It is also possible to modify model estimation settings prior to running the model by selecting Other Settings from the main menu bar, then Estimation Settings, as illustrated in Figure 3.8. Estimation settings were discussed in Chapter 2. For the null model, we use the default setting, restricted maximum likelihood estimation. There is also an Iterations Settings window, also from the Other Settings menu. Though not illustrated here, it provides options discussed in Chapter 2 with regard to estimation settings. One may also select Other Settings from the main menu bar, then Output Settings to obtain the window shown in Figure 3.9. For this model, one may choose to print out variance covariance matrices or to restrict output to the main results. The default is restricted output and no matrices. To run the null model, the researcher simply selects Run Analysis from the main menu bar. Output is sent to the file location and a name is specified in the Basic Model Specifications window (Figure 3.7). To view the output, select File, View Output, from the main menu bar. For this example, the critical output of the null model looks as shown in Table 3.2. The phrase Number of estimated parameters = 2 refers to the fact that in a null model, estimates are made for the level 1 intercept and the level 2 intercept. In the
10 64 PART I. GUIDE Figure 3.7 Basic Model Specifications window from Basic Settings menu in HLM 7 final variance components table, the fact that the component for the intercept (161.94, which HLM labels tau, τ) is significant means that the intercept of the outcome variable, SCORE0, is significantly affected by its predictors, which in this example is the level 2 effect of agency. A nonsignificant intercept in the variance components table term (not the case here) would mean that after other variables in the model are controlled, there would be no residual betweengroups variance in the level 1 dependent variable (Score0). The agency effect is smaller than the residual variance component (212.69, which HLM also labels sigmasquared, σ 2 ), indicating that there is still considerable residual variation in Score0 yet to be explained and that a model with additional predictors may be needed. The fact that the intercept component is significant means that the intraclass correlation coefficient, ICC, is also significant, indicating that a multilevel model is appropriate and needed. ICC varies from +1.0 when group means differ but within any group there is no variation, to  1/(n  1) when group means are all the same but withingroup variation is very large. At the extreme, when ICC
11 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 65 Figure 3.8 Estimation Settings window from Other Settings menu in HLM 7 Figure 3.9 Output Settings window for HLM2 Table 3.2 Final Estimation of Variance Components for the TwoLevel Null Model Random effect Standard deviation Variance component d.f. χ 2 pvalue INTRCPT1, u <0.001 level1, r Statistics for current covariance components model Deviance = Number of estimated parameters = 2
12 66 PART I. GUIDE approaches 0 or is negative, hierarchical modeling is not appropriate. For this example, the magnitude of ICC may be calculated as the intercept variance component in the null model divided by the total of variance components. That is, ICC = /( ) =.43. The fixed effect tables are of lesser interest in a null model but are presented in Table 3.3. Mean performance score (the intercept at level 1) is estimated to be for this example, when the level 2 grouping variable, agency, is the only effect modeled. Confidence limits around the mean, of course, are approximately plus or minus two standard errors. The lower table, with robust standard errors, produces the same estimate but has a slightly different standard error. Robust standard errors are recommended when it is possible the researcher has specified the wrong distribution of the dependent variable. Significant differences between the ordinary and robust estimates of the standard error may flag a problem with the distribution specified by the researcher. This is not the case in this example, which specified a normal distribution (which is the default). The Deviance value of in Table 3.2 is the basis of model fit measures. While not used at this point, for the null model it is the baseline model fit. More complex models are assessed in part by how greatly they reduce deviance (which is also called 2 log likelihood, 2LL, and model chisquare). These tests of the difference in deviance values between models are likelihood ratio tests, requested in HLM 7 by selecting Other Settings from the main menu bar, then Hypothesis Testing, as discussed later in this chapter. In summary, at the end of analysis of the null model we have demonstrated that there is a significant agency effect on employee performance scores; that therefore multilevel modeling is Table 3.3 Fixed Effects Tables for the Null Model Final estimation of fixed effects Fixed effect Coefficient Standard error tratio Approx.d.f. pvalue For INTRCPT1, β 0 INTRCPT2, γ <0.001 Final estimation of fixed effects (with robust standard errors) Fixed effect Coefficient Standard error tratio Approx. d.f. pvalue For INTRCPT1, β 0 INTRCPT2, γ <0.001
13 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 67 needed; and that additional, more complex models with more predictors should reduce significantly the baseline deviance value of A RANDOM COEFFICIENTS REGRESSION MODEL IN HLM 7 Given level 1 representing employees, with performance score as an outcome (dependent) variable, and level 2 representing agencies, a random coefficients regression model is one with one or more level 1 predictors such as gender, years of experience, or a binary indicator for whether the employee is certified or not. The level 2 grouping variable (Agency) remains a random factor, but there are no other level 2 predictors. The coefficients term in the label means that the agency effect is used not only to model the level 1 intercept of SCORE0 as an outcome, but also to model the regression coefficients of the level 1 predictors. As an example of random coefficients (RC) regression, employee performance score (score0) at level 1 is predicted from the level 1 covariates years of experience (YrsExper) and sex (Gender, where 0 = male, 1 = female). Note that HLM 7 enters binary variables like Gender as covariates by default. There are no predictors at level 2, but Agency is the subjects variable under which employees are grouped. The intercept of score0 at level 1 and the b coefficient of YrsExper at level 1 are both modeled as random effects of Agency. Gender is treated as a simple level 1 fixed effect. This model explores whether the Agency effect discovered in the null model may be attributed in part to some agencies having more experienced employees than others. The model also explores whether the demographic variable, Gender, modifies the relationship of years of experience to performance score. Figure 3.10 illustrates this RC regression model. An oftencited advantage of HLM software is how its user interface clearly separates regression models at different levels. Here, at level 1, score0 is predicted from YrsExper and Gender, plus an intercept term β 0j and an error term r ij : SCORE0 ij = β 0j + β 1j *(YRSEXPER ij ) + β 2j *(GENDER ij ) + r ij At level 2, there are no predictors. However, the level 1 intercept is predicted by the level 2 mean (γ 00 ) of score0 plus a level 2 error term (u 0j ). The level 2 error term represents the random effect of agency on score0 at level 1. Also, the level 1 regression coefficient (slope) of YrsExper (β 1j ) is predicted by the mean of agency regression coefficients where this mean is based on the 132 agencies in the sample, plus a level 2 error term (u 1j ) representing the random effect of Agency on the level 1 regression of score0 on YrsExper. HLM 7 output represents the level 2 equations as below: β 0j = γ 00 + u 0j β 1j = γ 10 + u 1j β 2j = γ 20
14 68 PART I. GUIDE The level 1 and level 2 equations can be combined, through substitution, into the single mixed model equation below: SCORE0 ij = γ 00 + γ 10 *YRSEXPER ij + γ 20 *GENDER ij + u 0j + u 1j *YRSEXPER ij + r ij Figure 3.10 A twolevel random coefficients regression model in HLM 7 As explained in Chapter 1, by designating a level 2 subjects variable, Agency, one is requesting that regressions be created separately for each agency so that the variance in intercepts and coefficients can be calculated and used in subsequent estimates of fixed and random effects. If full rather than reduced output is requested from the Other Settings, Output Settings menu selection, HLM 7 will print the OLS regression coefficients for the first 10 agencies (the default of 10 can be overridden by the researcher to get all coefficients), as shown in Table 3.4. Using these intercepts and slopes, one can create a plot of multiple different regression lines across agencies, graphically illustrating the nature of random intercepts and random slopes for the given data. The likelihood ratio test, discussed in Chapter 2, can be used as an overall test of whether the RC regression model with predictors is a significantly better fit than the interceptonly (null) model without predictors. In HLM 7, this is done from the main menu by selecting Other Settings, Hypothesis Testing, leading to the Hypothesis Testing window shown in Figure In this window, one
15 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 69 Table 3.4 OLS Coefficients for the First 10 Agencies in the RC Regression Model Level1 OLS Regressions Level 2 Unit INTRCPT1 YRSEXPER slope The average OLS level 1 coefficient for INTRCPT1 = The average OLS level 1 coefficient for YRSEXPER = Figure 3.11 Hypothesis Testing window in HLM 7
16 70 PART I. GUIDE manually enters the deviance value and number HLM 7 Likelihood Ratio Test for the of estimated parameters from the null model. Table 3.5 RC Regression Model Compared to The greater the drop in the deviance (2LL), the the Null Model more likely the fit is to be significantly better. For this example, deviance dropped from Deviance in the null model to in the Deviance = RC regression model. The likelihood ratio test, shown in HLM 7 output in Table 3.5, shows this Number of estimated parameters = 4 difference to be significant at better than the VarianceCovariance components test.001 level. Later, the likelihood ratio test can be χ 2 statistic = used to compare any two models if one is nested Degrees of freedom = 2 within the other, as the null model is nested within the RC regression model. pvalue = < Another way to assess improvement in model fit is to examine the residual variance component in the Final Estimation of Variance Components table (Table 3.6). The residual variance component is variance associated with the withinagency variation in score0 not accounted for by the random effects of Agency on the intercept of score0 and on the slope of YrsExper. As the random effects explain more, the residual component will drop. HLM 7 lists the residual variance component as level1, r. For these data, the residual component drops from in the null model to in the RC regression model. Since both were models in which the offdiagonal covariances were constrained to zero, 4 we can calculate that residual variance was /( ) = 46% of total variance in the RC regression model compared by similar calculation to 57% in the null model. In the same manner, in Table 3.6 we can calculate that the random effect of Agency on mean performance scores by agency (the INTRCPT1, u 0 effect) is 37% of total effects. This means that there is a significant, moderately strong Table 3.6 Variance Components Table for the RC Regression Model Final estimation of variance components Random effect Standard deviation Variance component d.f. χ 2 pvalue INTRCPT1, u <0.001 YRSEXPER slope, u <0.001 level1, r
17 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 71 tendency for some agencies to have higher mean scores than others. The random effect of Agency on slopes of YrsExper (the YRSEXPER slope, u 1 effect) is 18% of total effects. This means that there is a significant but weaker tendency for YrsExper to have a stronger effect on performance score in some agencies than others. Table 3.7 shows the fixed effects table for the RC regression model. HLM 7 prints two fixed effects tables, one using robust standard errors and one not. Robust standard errors should be used when it is possible that the distribution of the dependent variable was misspecified, though for this example the coefficients are identical. For these data, score0 is normally distributed, as specified (for example, skew and kurtosis were both within +/ 1.0), and coefficients and the probability levels are identical between the two versions of the fixed effects table. From Table 3.7, we conclude that both YrsExper and Gender are significant predictors of employee performance scores (score0). YrsExper was centered (mean = 0) and for Gender male was coded 0. Therefore, a male with mean years of experience could be expected to score 59.4 points on the performance test. The fact that the slope of Gender was negative means that being female (Gender = 1) was associated with scoring 6.5 points less, controlling for other variables in the model. The tratios are the regression coefficients divided by their standard errors. The fact that the absolute tratios for YrsExper and Gender are similar indicates similarity in effect size. (Note that a 1year increase in years of experience does not predict an increase of score of points because YrsExper was not only centered but was also transformed to decimal form, ranging from .79 to +.56: That is, the units were no longer raw years.) Table 3.7 The Fixed Effects Table for the RC Regression Model Final estimation of fixed effects (with robust standard errors) Fixed effect Coefficient Standard error tratio Approx. d.f. pvalue For INTRCPT1, β 0 INTRCPT2, γ <0.001 For YRSEXPER slope, β 1 INTRCPT2, γ <0.001 For GENDER slope, β 2 INTRCPT2, γ <0.001
18 72 PART I. GUIDE HOMOGENOUS AND HETEROGENEOUS FULL RANDOM COEFFICIENTS MODELS Also called an interceptsandslopesasoutcomes model, the full random coefficients model is a type of hierarchical linear model in which, for two levels, there are predictors at both levels, and both the level 1 intercept and the level 1 slopes are predicted as random effects. As shown in Figure 3.12 below, the level 1 model for this example remains similar to the previous model. The essential features of the model are these: At level 1, Score0 is predicted by YrsExper and Gender. Agency remains the grouping variable defining level 2, which means that as many level 1 regressions are run as there are agencies in the sample, yielding an estimate of the variability of level 1 slopes and intercepts. The intercept at level 1 is predicted as a random effect of Agency and of the level 2 predictor AgencyPC, which is a newly added covariate measuring percent of employees certified in an agency. The slope of YrsExper at level 1 is predicted as a random effect of Agency. In a subsequent example later in this section, a different model will illustrate modeling the slope of YrsExper as a random effect of both Agency and AgencyPC at level 2. The current model tests the proposition that the mean (intercept) for Score0 is a random effect of Agency and of AgencyPC at level 2, and that the strength of relationship (slope) of YrsExper to Score0 is a function of Agency but not of AgencyPC. The slope of Gender is not predicted as a random effect, as signified by its random error term (u2) being grayed out in Figure Figure 3.12 A twolevel random coefficients model in HLM 7
19 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 73 Though not illustrated, we click Other Settings to obtain a window similar to that shown in Figure 3.11 above. We enter the deviance ( ) and number of parameters (4) from the previous RC regression model without the level 2 agency percent certified variable. This requests a likelihood ratio test of the model fit difference, to be discussed below, between the current model and the previous one. In a second run, a likelihood test was also requested comparing the current model with the null model. In addition, the Test level1 homogeneity of variance checkbox, also illustrated in Figure 3.11 above, was checked. (Figure 3.11 also shows the HLM 7 option for Multivariate Hypotheses Tests, as illustrated, for example, in Chapter 12.) The overall test of the model that is reflected in the likelihood ratio test of the difference in deviance (2LL) between models is shown in Table 3.8. While deviance declined another points Table 3.8 HLM 7 Likelihood Ratio Test for the Two Level Homogenous Full RC Model Statistics for current covariance components model Deviance = Number of estimated parameters = 4 VarianceCovariance components test (compared to RC regression model) χ 2 statistic = Degrees of freedom = 0 pvalue = >.500 Table 3.9 Test of Homogeneity of Level 1 Variance Test of homogeneity of level 1 variance χ 2 statistic = Degrees of freedom = 131 pvalue = compared to the RC regression model and reflects significantly better fit than the null model, the difference between the current model and the RC regression model is not significant (pvalue = >.500). That is, agency percent certified as a level 2 covariate in the current model reduced deviance by a nonsignificant amount. On parsimony grounds, the researcher would prefer the RC regression model. Nonetheless, for instructional purposes, the remaining HLM 7 output is examined below. The Test level1 homogeneity of variance output, not previously discussed, is shown in Table 3.9. Although not invoking this test is the default in HLM 7, it is sufficiently critical that Raudenbush and Bryk (2002) state, investigators generally will wish to begin with this assumption (p. 263). This test refers to the assumption that when the model is run for each of the 132 agencies in the current example, the residual variances are homogenous. Optionally, the researcher may override the default and test this assumption. For the example data, the test pvalue is This finding of significance means that residual variances do differ significantly for these data across agencies. Raudenbush and Bryk (p. 263) note that heterogeneity of error variance is a serious problem if variances are not random but are a function of
20 74 PART I. GUIDE level 1 or level 2 predictors. Heterogeneity may indicate one of four problems in the research design: 1. Model misspecification. One or more level 1 predictors have been omitted from the model, where the variables in question are distributed with unequal variance across groups. This is the most likely cause of heterogeneity, making this test a form of screening for model misspecification. 2. A level 1 predictor has been modeled as a fixed effect when in fact it is a random effect. 3. One or more level 1 predictors are nonnormal (for example, kurtotic with heavy tails), causing the significance test statistic for homogeneity of residual variance, which assumes normality, to report lack of homogeneity. Transformation of the predictor, as in OLS regression, may mitigate nonnormality. Likewise, use of a link function other than identity in a generalized linear mixed model may also moderate the effects of nonnormality. 4. Coding or other data entry errors, or presence of outliers, may cause heterogeneous error variance in some groups. It is possible to visually inspect residual variance by level 2 groups (agencies for this example) in a variety of ways (see Raudenbush & Bryk, 2002, pp ; Raudenbush et al., 2011, pp ). For instance, within HLM 7 one may select File, Graph Equations, Level 1 residual versus predicted value to obtain a plot where points represent agencies. A more complex graphical method is to use a combination of HLM 7, SPSS, and Excel, yielding a plot of residual variances by agency, as shown in Figure The process is described in the endnotes to this chapter. 5 The steeper the trend line and the more outliers in this chart, the less homogenous residual variance is. Sometimes it may be possible to obtain a nonsignificant test of homogeneity by identifying outliers, such as Agency 48 in Figure 3.13, and removing them from the sample for separate analysis. However, if the trend line is steep, the researcher may need to create a heterogeneous model. Endnote 5 describes how to obtain a table of the correlations of Agency residual variance with the predictor variables, showing for these data that such variance is correlated with years of experience though not with gender. To deal with the problem of heterogeneous residual variance, the four possibilities above need to be explored first. Ultimately, however, it may be necessary to create a heterogeneous variance model. This is done in HLM 7 by selecting Other Settings, Estimation Settings, giving the window shown in Figure 3.8 above. Clicking the Heterogeneous sigma^2 button leads to the dialog screen of the same name, shown in Figure Here, one must enter a level 1 variable, which may account for heterogeneous residual variance. The variable may be one not otherwise used in the model.
21 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 75 Figure 3.13 Residual variance by agency for the twolevel random coefficients model Figure 3.14 Modeling heterogeneity in HLM 7
22 76 PART I. GUIDE Determining which variable to use as the predictor of level 1 residual variance can be explored using a statistical package to view the variances of the OLS residuals by candidate variables. In this example, in SPSS, select Analyze, Regression, Linear; set the dependent to Score0. Set the independents to YrsExper and Gender, as in the current model. Set the Selection variable to be Certified (coded 0 = not certified, 1 = certified). Set the Rule to be Certified = 0 in a first run and Certified = 1 in a second run. In the ANOVA table output, residual sum of squares for not certified is 57,488.9 and for certified is 668, The larger the difference, the better the selection variable is as a candidate predictor to be specified in Figure 3.14 when creating a heterogeneous variance model. Having determined that the level 1 variable Likelihood Ratio Test and Certified is a good candidate to predict residual Homogeneity of Residual Variance variance, Certified could be incorporated in the Table 3.10 Test for the Heterogeneous Full model in one of three ways: (1) It could be Random Coefficients Model added as a level 1 variable not modeled as a random effect of level 2; (2) it could be added and also modeled as a random effect; or (3) it Statistics for the current model could be used as the predictor of level 1 residual Deviance = variance by using a heterogeneous residual variance model. For these data, option (1) would Number of estimated parameters = 9 yield results that fail the test of level 1 homogeneity of variance and that fail to show signifi Model comparison test cantly better fit than the previous RC regression χ 2 statistic = model. Option (2) would drop the deviance Degrees of freedom = 5 value enough to show significantly better fit but pvalue = <0.001 would also fail the test of level 1 homogeneity of residual variance. As shown in Table 3.10, Test of homogeneity of level 1 variance option 3, the heterogeneous model, yields results χ 2 statistic = that pass the homogeneity test (the homogeneity Degrees of freedom = 131 pvalue is nonsignificant) and that also show pvalue = significantly better fit than the RC regression model discussed earlier (the model comparison likelihood ratio test pvalue is significant). As shown in Figure 3.15, the heterogeneous model is identical to that shown in Figure 3.12 except that at level 1, terms are added that model residual variance as a function of the predictor variable Certified. Fixed effects of the heterogeneous model are shown in Table HLM prints out two fixed effects tables, one with robust standard errors and one with ordinary standard errors. Robust standard errors are advisable when there is misspecification of the distribution of the dependent variable. Therefore, significant differences between the ordinary and robust estimates of the standard error may flag a problem with the distribution specified by the researcher (normal is default). This does not appear to be a problem for this example, and
23 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 77 Figure 3.15 The heterogeneous random coefficients model in HLM 7 Table 3.11 Fixed Effects for the Heterogeneous Full Random Coefficients Model Final estimation of fixed effects (with robust standard errors) Fixed effect Coefficient Standard error tratio Approx. d.f. pvalue For INTRCPT1, β 0 INTRCPT2, γ <0.001 AGENCYPC, γ <0.001 For YRSEXPER slope, β 1 INTRCPT2, γ <0.001 For GENDER slope, β 2 INTRCPT2, γ <0.001 one arrives at the same statistical inferences either way. For this example, all fixed effects are significant: The level 1 intercept of 56.5 gives the mean performance score across all agencies when other variables in the model are controlled at zero. Since YrsExper and
24 78 PART I. GUIDE AgencyPC were centered, controlling means when both are at their mean values. Since Gender was coded such that 0 = male, controlling means for men. That is, men with mean years of experience working in agencies with a mean percent of certified employees are predicted to have performance scores, on the average, of The level 2 predictor AgencyPC (percent certified for a given agency) is significant. This variable, which ranged from.11 to 1.0, was centered when used in the HLM model. Its coefficient is positive, meaning that the higher the percent certified in an agency, the higher the intercept (hence the higher the mean performance score) for employees of the agency. Its slope, 19.8, means that when the percent certified goes up by 1 percent (which is.01 units on a scale from.11 to 1.0), mean performance scores are expected to rise by.198 units. The slope of the level 1 predictor variable Gender is significant and negative. Since Gender was coded 0 = male, 1 = female, a 1unit increase means being female. That is, being female lowered the expected performance score by 4.79 points, on the average, controlling for other variables in the model. The slope of the level 1 predictor YrsExper was significant and positive. This variable was centered on its mean and expressed in standardized units ranging from .79 to +.56 (not in raw years). The fact that it is positive means the more years experience, the higher the predicted performance score, controlling other variables in the model. That its tratio in absolute terms is about twice that for Gender means YrsExper has a greater effect on performance score than does Gender. Table 3.12 shows the random effects components in the heterogeneous random coefficients model, of which there are two. Because residual variance is being modeled in a heterogeneous model, there is no residual variance component as, for instance, shown in Table 3.6 for a related homogenous model. The Intrcpt1 component is the effect on mean performance scores at level 1 due to Agency as a random factor at level 2. The YrsExper slope component is the effect on the slope of YrsExper and level 1 due to Agency as a random factor at level 2. Both effects are significant. Thus, both mean performance score and the regression coefficients for YrsExper vary significantly by Agency, confirming that multilevel analysis is required to properly model employee performance score at level 1. Coefficients in the fixed effects table discussed above will be more reliable than coefficients from the equivalent model in OLS regression due to the existence of these significant random effects. That the intercept component Table 3.12 Random Effects for the Heterogeneous Random Coefficients Model Final estimation of variance components Random effect Standard deviation Variance component d.f. χ 2 pvalue INTRCPT1, u <0.001 YRSEXPER slope, u <0.001
25 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 79 is significant even controlling for other variables in the model means that there remains significant variation, which might be explained by adding additional predictors to the model. As a final example in this section, Figure 3.16 illustrates a full random coefficients model in HLM otherwise paralleling the foregoing one, but now with the level 2 predictor, centered agency percent certified (AgencyPC), used to model the level 1 slope of years of experience (YrsExper) as well as the level 1 intercept. As can be seen in Figure 3.16, this is simply a matter of adding AgencyPC to the level 2 equation that models the level 1 slope of YrsExper. In the level 2 equation, β 1j = γ 10 + γ 11 *(AGENCYPC j ) + u 1j, the u 1j term is the Agency effect on the slope of YrsExper and the γ 11 *(AGENCYPC j ) term is the AgencyPC effect on the slope. Table 3.13 shows the deviance model fit statistic for this model. It is very close to the homogenous RC model in which AgencyPC modeled only the level 1 intercept and not the slope (Table 3.8). This strongly hints at the finding shown below in the fixed effects table, which shows the AgencyPC effect on the slope of YrsExper to be nonsignificant (Table 3.14). We conclude that agency percent certified at level 2 does not account for a significant portion of the variance in the strengths (slopes) of the relation of YrsExper with performance scores across Figure 3.16 A full random coefficients model with a level 2 covariate modeling level 1 slope and intercept
26 80 PART I. GUIDE agencies. Neither of these homogenous Deviance and Likelihood Ratio Test of models has as good a fit as did the heterogeneous model just discussed. The likelihood the Homogenous RC Model With Agency Table 3.13 Modeling Both Slope and Intercept, ratio tests compare the full homogenous Compared to the RC Regression Model RC model (Table 3.8), the full heterogeneous RC model (Table 3.10), and the full Statistics for current covariance components model homogenous RC model with agency percent certified modeling slope as well as Deviance = Number of estimated parameters = 4 intercept. (Table 3.13 shows only the heterogeneous model to be better fitting than VarianceCovariance components test the RC regression model with no level 2 χ predictors.) 2 statistic = Because the interpretation parallels that Degrees of freedom = 0 of full RC models discussed above, the pvalue = >.500 variance components table for the homogenous RC model with agency percent certified modeling slope as well as intercept is not presented. Also, though not presented in table form, it may be noted that this homogenous full RC model just discussed fails the test of homogeneity of level 1 variance, as did the earlier homogenous full RC model. However, for pedagogic reasons and because SAS and SPSS cannot easily compute models for heterogeneous error variance, this is the model reproduced in SAS 9.2 in Chapter 4 and in SPSS 19 in Chapter 5. It should be noted that such testing for homogeneity of error variance routinely returns a finding of heterogeneity, yet it is common practice to execute homogenous models anyway, in part because the same substantive Table 3.14 Fixed Effects for the Homogenous RC Model With AGENCYPC Modeling Both Slope and Intercept Final estimation of fixed effects (with robust standard errors) Fixed effect Coefficient Standard error tratio Approx. d.f. pvalue For INTRCPT1, β 0 INTRCPT2, γ < AGENCYPC, γ < For YRSEXPER slope, β 1 INTRCPT2, γ < AGENCYPC, γ For GENDER slope, β 2 INTRCPT2, γ < 0.001
27 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 81 conclusions are often arrived at. Also, even if error variance is heterogeneous, it may not be correlated with predictor variables (though it is correlated with YrsExper for these data). How to check is discussed in the endnotes. 6 Figure 3.17 Correlations of residual error variance with level 1 predictors THREELEVEL HIERARCHICAL LINEAR MODELS Threelevel problems occur frequently in realworld data: crosssectional studies of students nested within classrooms nested within schools, for instance, or longitudinal studies of yearly tests nested within students nested within schools. To create a basic, crosssectional threelevel hierarchical linear model, we will use the same SPSS data file, mixed3level.sav, which had within it a previously undiscussed field for the third level: Department (Employees are nested within Agencies, and Agencies within Departments, with Department being the department ID variable) and two departmentlevel covariates to serve as potential level 3 predictors. As HLM 7 has only minimal data management capabilities, like most HLM users, we initially create the dataset in another statistical package, in this case SPSS 19. SAS may also be used for this purpose, as HLM 7 reads both SPSS and SAS formats directly. While separate data files might be created for levels 1, 2, and 3, this is not necessary and was not done for the example data. The first step is to import the data from SPSS.sav format and to create the HLM 7 multivariate data matrix file (.mdm format) and data template file (.mdmt format). This assumes that the data are previously sorted in a nested order: employees within agencies within departments. The.mdm file will be incorrect if
28 82 PART I. GUIDE data are not sorted properly in SPSS or other statistical packages used for data management. Threelevel crosssectional models are created in the HLM3 module. When HLM 7 opens, select File, Make new MDM file, Stat package input. A Select MDM Type window opens. Check the HLM3 radio button and click OK to arrive at the Make MDM HLM3 window shown in Figure Figure 3.18 HLM 7 setup for threelevel crosssectional models (HLM3) 1. In the Make MDM HLM3 window, follow these steps: 2. In the upper right, first enter the desired filename for the data matrix file. Here, it is M3_3L.mdm. 3. In the Level1 Specification area, click the Browse button and browse to the SPSS.sav file and enter it. Also click the appropriate radio buttons regarding how to handle missing data, if any. 4a. Click the Choose Variables button to enter level 1 variables as shown in Figure For level 1, enter all three ID variables (Department, Agency, and Employee)
29 CHAPTER 3. Introductory Guide to HLM With HLM 7 Software 83 Figure 3.19 Variable selection for threelevel models in HLM 7 as shown, as well as entering any level 1 covariates and the level 1 dependent variable in in MDM checkboxes. For level 2, enter just Department and Agency, plus any level 2 predictors. For level 3, enter just Department, plus any level 3 predictors. 4b. Repeat for levels 2 and Save the template file. 6. Make (and save) the data matrix file. 7. Click Check Statistics to get a window showing the count of cases and descriptive statistics (mean, standard deviation, minimum, maximum) at each level. While this is an optional step, it is wise to check to see if data are being imported as expected. 8. Click Done to finish, and exit to the WHLM: hlm3file window, from which one may select options and run a threelevel model. Alternatively, one may close and later run HLM 7, then select File, Create a new model from an existing.mdm file. After the.mdm file is created, any of a variety of different crosssectional threelevel models may be created. With any model, the first analysis step is to specify the distribution of the dependent variable. For the example data, the dependent is employee performance score (Score0), which is normally distributed. This is the default selection, shown earlier in Figure 3.7. In the same Basic
30 84 PART I. GUIDE Model Specifications window, the researcher should also specify the file location for the output and the title for the given run of the model. Model A Figure 3.20 shows the threelevel null model in HLM 7 (compare to the twolevel null model in Figure 3.6). For the null model, there are no predictors at any level. However, the intercept of performance score at level 1 is modeled for the Agency grouping effect at level 2 (signified by the r 0 random effects term in the level 2 equation). There is also a Department grouping effect at level 3 (signified by the u 00 random effects term). The key element of the output is that the deviance is , a value that can be compared with later models with predictors. Also, both the level 2 and level 3 intercepts were significant, confirming the existence of both a level 2 Agency effect and a level 3 Department effect. Finally, the residual withingroup variance component was , a baseline value that will decrease as predictors are added to the model. Figure 3.20 The threelevel null model in HLM 7 (Model A)
Introduction to Multilevel Modeling Using HLM 6. By ATS Statistical Consulting Group
Introduction to Multilevel Modeling Using HLM 6 By ATS Statistical Consulting Group Multilevel data structure Students nested within schools Children nested within families Respondents nested within interviewers
More informationIntroducing the Multilevel Model for Change
Department of Psychology and Human Development Vanderbilt University GCM, 2010 1 Multilevel Modeling  A Brief Introduction 2 3 4 5 Introduction In this lecture, we introduce the multilevel model for change.
More informationΕισαγωγή στην πολυεπίπεδη μοντελοποίηση δεδομένων με το HLM. Βασίλης Παυλόπουλος Τμήμα Ψυχολογίας, Πανεπιστήμιο Αθηνών
Εισαγωγή στην πολυεπίπεδη μοντελοποίηση δεδομένων με το HLM Βασίλης Παυλόπουλος Τμήμα Ψυχολογίας, Πανεπιστήμιο Αθηνών Το υλικό αυτό προέρχεται από workshop που οργανώθηκε σε θερινό σχολείο της Ευρωπαϊκής
More informationUse of deviance statistics for comparing models
A likelihoodratio test can be used under full ML. The use of such a test is a quite general principle for statistical testing. In hierarchical linear models, the deviance test is mostly used for multiparameter
More informationIntroduction to Data Analysis in Hierarchical Linear Models
Introduction to Data Analysis in Hierarchical Linear Models April 20, 2007 Noah Shamosh & Frank Farach Social Sciences StatLab Yale University Scope & Prerequisites Strong applied emphasis Focus on HLM
More informationCHAPTER 9 EXAMPLES: MULTILEVEL MODELING WITH COMPLEX SURVEY DATA
Examples: Multilevel Modeling With Complex Survey Data CHAPTER 9 EXAMPLES: MULTILEVEL MODELING WITH COMPLEX SURVEY DATA Complex survey data refers to data obtained by stratification, cluster sampling and/or
More informationNCSS Statistical Software Principal Components Regression. In ordinary least squares, the regression coefficients are estimated using the formula ( )
Chapter 340 Principal Components Regression Introduction is a technique for analyzing multiple regression data that suffer from multicollinearity. When multicollinearity occurs, least squares estimates
More informationSTATISTICA Formula Guide: Logistic Regression. Table of Contents
: Table of Contents... 1 Overview of Model... 1 Dispersion... 2 Parameterization... 3 SigmaRestricted Model... 3 Overparameterized Model... 4 Reference Coding... 4 Model Summary (Summary Tab)... 5 Summary
More informationAn introduction to hierarchical linear modeling
Tutorials in Quantitative Methods for Psychology 2012, Vol. 8(1), p. 5269. An introduction to hierarchical linear modeling Heather Woltman, Andrea Feldstain, J. Christine MacKay, Meredith Rocchi University
More informationConstruct the MDM file for a HLM2 model using SPSS file input
Construct the MDM file for a HLM2 model using SPSS file input Level1 file: For HS&B example data (distributed with the program), the level1 file (hsb1.sav) has 7,185 cases and four variables (not including
More informationData analysis process
Data analysis process Data collection and preparation Collect data Prepare codebook Set up structure of data Enter data Screen data for errors Exploration of data Descriptive Statistics Graphs Analysis
More informationOverview Classes. 123 Logistic regression (5) 193 Building and applying logistic regression (6) 263 Generalizations of logistic regression (7)
Overview Classes 123 Logistic regression (5) 193 Building and applying logistic regression (6) 263 Generalizations of logistic regression (7) 24 Loglinear models (8) 54 1517 hrs; 5B02 Building and
More informationIntroduction to Longitudinal Data Analysis
Introduction to Longitudinal Data Analysis Longitudinal Data Analysis Workshop Section 1 University of Georgia: Institute for Interdisciplinary Research in Education and Human Development Section 1: Introduction
More informationDirections for using SPSS
Directions for using SPSS Table of Contents Connecting and Working with Files 1. Accessing SPSS... 2 2. Transferring Files to N:\drive or your computer... 3 3. Importing Data from Another File Format...
More informationIntroduction to Regression and Data Analysis
Statlab Workshop Introduction to Regression and Data Analysis with Dan Campbell and Sherlock Campbell October 28, 2008 I. The basics A. Types of variables Your variables may take several forms, and it
More informationLogistic Regression. http://faculty.chass.ncsu.edu/garson/pa765/logistic.htm#sigtests
Logistic Regression http://faculty.chass.ncsu.edu/garson/pa765/logistic.htm#sigtests Overview Binary (or binomial) logistic regression is a form of regression which is used when the dependent is a dichotomy
More informationChapter 5 Analysis of variance SPSS Analysis of variance
Chapter 5 Analysis of variance SPSS Analysis of variance Data file used: gss.sav How to get there: Analyze Compare Means Oneway ANOVA To test the null hypothesis that several population means are equal,
More informationThe 3Level HLM Model
James H. Steiger Department of Psychology and Human Development Vanderbilt University Regression Modeling, 2009 1 2 Basic Characteristics of the 3level Model Level1 Model Level2 Model Level3 Model
More informationData Analysis Tools. Tools for Summarizing Data
Data Analysis Tools This section of the notes is meant to introduce you to many of the tools that are provided by Excel under the Tools/Data Analysis menu item. If your computer does not have that tool
More information11. Analysis of Casecontrol Studies Logistic Regression
Research methods II 113 11. Analysis of Casecontrol Studies Logistic Regression This chapter builds upon and further develops the concepts and strategies described in Ch.6 of Mother and Child Health:
More informationSpecifications for this HLM2 run
One way ANOVA model 1. How much do U.S. high schools vary in their mean mathematics achievement? 2. What is the reliability of each school s sample mean as an estimate of its true population mean? 3. Do
More informationJanuary 26, 2009 The Faculty Center for Teaching and Learning
THE BASICS OF DATA MANAGEMENT AND ANALYSIS A USER GUIDE January 26, 2009 The Faculty Center for Teaching and Learning THE BASICS OF DATA MANAGEMENT AND ANALYSIS Table of Contents Table of Contents... i
More informationProfile analysis is the multivariate equivalent of repeated measures or mixed ANOVA. Profile analysis is most commonly used in two cases:
Profile Analysis Introduction Profile analysis is the multivariate equivalent of repeated measures or mixed ANOVA. Profile analysis is most commonly used in two cases: ) Comparing the same dependent variables
More informationWhen to use Excel. When NOT to use Excel 9/24/2014
Analyzing Quantitative Assessment Data with Excel October 2, 2014 Jeremy Penn, Ph.D. Director When to use Excel You want to quickly summarize or analyze your assessment data You want to create basic visual
More informationSAS Software to Fit the Generalized Linear Model
SAS Software to Fit the Generalized Linear Model Gordon Johnston, SAS Institute Inc., Cary, NC Abstract In recent years, the class of generalized linear models has gained popularity as a statistical modeling
More informationThis chapter will demonstrate how to perform multiple linear regression with IBM SPSS
CHAPTER 7B Multiple Regression: Statistical Methods Using IBM SPSS This chapter will demonstrate how to perform multiple linear regression with IBM SPSS first using the standard method and then using the
More informationIllustration (and the use of HLM)
Illustration (and the use of HLM) Chapter 4 1 Measurement Incorporated HLM Workshop The Illustration Data Now we cover the example. In doing so we does the use of the software HLM. In addition, we will
More informationThe Basic TwoLevel Regression Model
2 The Basic TwoLevel Regression Model The multilevel regression model has become known in the research literature under a variety of names, such as random coefficient model (de Leeuw & Kreft, 1986; Longford,
More informationAnalyzing Intervention Effects: Multilevel & Other Approaches. Simplest Intervention Design. Better Design: Have Pretest
Analyzing Intervention Effects: Multilevel & Other Approaches Joop Hox Methodology & Statistics, Utrecht Simplest Intervention Design R X Y E Random assignment Experimental + Control group Analysis: t
More informationBill Burton Albert Einstein College of Medicine william.burton@einstein.yu.edu April 28, 2014 EERS: Managing the Tension Between Rigor and Resources 1
Bill Burton Albert Einstein College of Medicine william.burton@einstein.yu.edu April 28, 2014 EERS: Managing the Tension Between Rigor and Resources 1 Calculate counts, means, and standard deviations Produce
More informationHYPOTHESIS TESTING: CONFIDENCE INTERVALS, TTESTS, ANOVAS, AND REGRESSION
HYPOTHESIS TESTING: CONFIDENCE INTERVALS, TTESTS, ANOVAS, AND REGRESSION HOD 2990 10 November 2010 Lecture Background This is a lightning speed summary of introductory statistical methods for senior undergraduate
More informationChapter 7: Simple linear regression Learning Objectives
Chapter 7: Simple linear regression Learning Objectives Reading: Section 7.1 of OpenIntro Statistics Video: Correlation vs. causation, YouTube (2:19) Video: Intro to Linear Regression, YouTube (5:18) 
More informationSPSS TRAINING SESSION 3 ADVANCED TOPICS (PASW STATISTICS 17.0) Sun Li Centre for Academic Computing lsun@smu.edu.sg
SPSS TRAINING SESSION 3 ADVANCED TOPICS (PASW STATISTICS 17.0) Sun Li Centre for Academic Computing lsun@smu.edu.sg IN SPSS SESSION 2, WE HAVE LEARNT: Elementary Data Analysis Group Comparison & Oneway
More informationHow to Get More Value from Your Survey Data
Technical report How to Get More Value from Your Survey Data Discover four advanced analysis techniques that make survey research more effective Table of contents Introduction..............................................................2
More informationAn analysis method for a quantitative outcome and two categorical explanatory variables.
Chapter 11 TwoWay ANOVA An analysis method for a quantitative outcome and two categorical explanatory variables. If an experiment has a quantitative outcome and two categorical explanatory variables that
More informationbusiness statistics using Excel OXFORD UNIVERSITY PRESS Glyn Davis & Branko Pecar
business statistics using Excel Glyn Davis & Branko Pecar OXFORD UNIVERSITY PRESS Detailed contents Introduction to Microsoft Excel 2003 Overview Learning Objectives 1.1 Introduction to Microsoft Excel
More informationModule 3: Correlation and Covariance
Using Statistical Data to Make Decisions Module 3: Correlation and Covariance Tom Ilvento Dr. Mugdim Pašiƒ University of Delaware Sarajevo Graduate School of Business O ften our interest in data analysis
More informationCOMPARISONS OF CUSTOMER LOYALTY: PUBLIC & PRIVATE INSURANCE COMPANIES.
277 CHAPTER VI COMPARISONS OF CUSTOMER LOYALTY: PUBLIC & PRIVATE INSURANCE COMPANIES. This chapter contains a full discussion of customer loyalty comparisons between private and public insurance companies
More informationSPSS Explore procedure
SPSS Explore procedure One useful function in SPSS is the Explore procedure, which will produce histograms, boxplots, stemandleaf plots and extensive descriptive statistics. To run the Explore procedure,
More informationModule 4  Multiple Logistic Regression
Module 4  Multiple Logistic Regression Objectives Understand the principles and theory underlying logistic regression Understand proportions, probabilities, odds, odds ratios, logits and exponents Be
More informationRegression Analysis: A Complete Example
Regression Analysis: A Complete Example This section works out an example that includes all the topics we have discussed so far in this chapter. A complete example of regression analysis. PhotoDisc, Inc./Getty
More informationDISCRIMINANT FUNCTION ANALYSIS (DA)
DISCRIMINANT FUNCTION ANALYSIS (DA) John Poulsen and Aaron French Key words: assumptions, further reading, computations, standardized coefficents, structure matrix, tests of signficance Introduction Discriminant
More informationSimple Linear Regression Inference
Simple Linear Regression Inference 1 Inference requirements The Normality assumption of the stochastic term e is needed for inference even if it is not a OLS requirement. Therefore we have: Interpretation
More informationSimple Predictive Analytics Curtis Seare
Using Excel to Solve Business Problems: Simple Predictive Analytics Curtis Seare Copyright: Vault Analytics July 2010 Contents Section I: Background Information Why use Predictive Analytics? How to use
More informationIBM SPSS Statistics 20 Part 4: ChiSquare and ANOVA
CALIFORNIA STATE UNIVERSITY, LOS ANGELES INFORMATION TECHNOLOGY SERVICES IBM SPSS Statistics 20 Part 4: ChiSquare and ANOVA Summer 2013, Version 2.0 Table of Contents Introduction...2 Downloading the
More information2013 MBA Jump Start Program. Statistics Module Part 3
2013 MBA Jump Start Program Module 1: Statistics Thomas Gilbert Part 3 Statistics Module Part 3 Hypothesis Testing (Inference) Regressions 2 1 Making an Investment Decision A researcher in your firm just
More informationStatistical Models in R
Statistical Models in R Some Examples Steven Buechler Department of Mathematics 276B Hurley Hall; 16233 Fall, 2007 Outline Statistical Models Structure of models in R Model Assessment (Part IA) Anova
More informationCHAPTER 3 EXAMPLES: REGRESSION AND PATH ANALYSIS
Examples: Regression And Path Analysis CHAPTER 3 EXAMPLES: REGRESSION AND PATH ANALYSIS Regression analysis with univariate or multivariate dependent variables is a standard procedure for modeling relationships
More informationKSTAT MINIMANUAL. Decision Sciences 434 Kellogg Graduate School of Management
KSTAT MINIMANUAL Decision Sciences 434 Kellogg Graduate School of Management Kstat is a set of macros added to Excel and it will enable you to do the statistics required for this course very easily. To
More informationAnalyzing Data from Small N Designs Using Multilevel Models: A Procedural Handbook
Analyzing Data from Small N Designs Using Multilevel Models: A Procedural Handbook Eden Nagler, M.Phil The Graduate Center, CUNY David Rindskopf, PhD CoPrincipal Investigator The Graduate Center, CUNY
More informationAssignments Analysis of Longitudinal data: a multilevel approach
Assignments Analysis of Longitudinal data: a multilevel approach Frans E.S. Tan Department of Methodology and Statistics University of Maastricht The Netherlands Maastricht, Jan 2007 Correspondence: Frans
More informationGeneralized Linear Models
Generalized Linear Models We have previously worked with regression models where the response variable is quantitative and normally distributed. Now we turn our attention to two types of models where the
More informationChapter 23. Inferences for Regression
Chapter 23. Inferences for Regression Topics covered in this chapter: Simple Linear Regression Simple Linear Regression Example 23.1: Crying and IQ The Problem: Infants who cry easily may be more easily
More informationChapter 7. Oneway ANOVA
Chapter 7 Oneway ANOVA Oneway ANOVA examines equality of population means for a quantitative outcome and a single categorical explanatory variable with any number of levels. The ttest of Chapter 6 looks
More informationMain Effects and Interactions
Main Effects & Interactions page 1 Main Effects and Interactions So far, we ve talked about studies in which there is just one independent variable, such as violence of television program. You might randomly
More informationUsing Excel for Statistics Tips and Warnings
Using Excel for Statistics Tips and Warnings November 2000 University of Reading Statistical Services Centre Biometrics Advisory and Support Service to DFID Contents 1. Introduction 3 1.1 Data Entry and
More informationGLM I An Introduction to Generalized Linear Models
GLM I An Introduction to Generalized Linear Models CAS Ratemaking and Product Management Seminar March 2009 Presented by: Tanya D. Havlicek, Actuarial Assistant 0 ANTITRUST Notice The Casualty Actuarial
More informationSPSS Introduction. Yi Li
SPSS Introduction Yi Li Note: The report is based on the websites below http://glimo.vub.ac.be/downloads/eng_spss_basic.pdf http://academic.udayton.edu/gregelvers/psy216/spss http://www.nursing.ucdenver.edu/pdf/factoranalysishowto.pdf
More informationRegression stepbystep using Microsoft Excel
Step 1: Regression stepbystep using Microsoft Excel Notes prepared by Pamela Peterson Drake, James Madison University Type the data into the spreadsheet The example used throughout this How to is a regression
More informationSPSS Guide: Regression Analysis
SPSS Guide: Regression Analysis I put this together to give you a stepbystep guide for replicating what we did in the computer lab. It should help you run the tests we covered. The best way to get familiar
More informationSection Format Day Begin End Building Rm# Instructor. 001 Lecture Tue 6:45 PM 8:40 PM Silver 401 Ballerini
NEW YORK UNIVERSITY ROBERT F. WAGNER GRADUATE SCHOOL OF PUBLIC SERVICE Course Syllabus Spring 2016 Statistical Methods for Public, Nonprofit, and Health Management Section Format Day Begin End Building
More informationANALYSING LIKERT SCALE/TYPE DATA, ORDINAL LOGISTIC REGRESSION EXAMPLE IN R.
ANALYSING LIKERT SCALE/TYPE DATA, ORDINAL LOGISTIC REGRESSION EXAMPLE IN R. 1. Motivation. Likert items are used to measure respondents attitudes to a particular question or statement. One must recall
More informationAnalysis of Variance. MINITAB User s Guide 2 31
3 Analysis of Variance Analysis of Variance Overview, 32 OneWay Analysis of Variance, 35 TwoWay Analysis of Variance, 311 Analysis of Means, 313 Overview of Balanced ANOVA and GLM, 318 Balanced
More informationModeration. Moderation
Stats  Moderation Moderation A moderator is a variable that specifies conditions under which a given predictor is related to an outcome. The moderator explains when a DV and IV are related. Moderation
More informationABSORBENCY OF PAPER TOWELS
ABSORBENCY OF PAPER TOWELS 15. Brief Version of the Case Study 15.1 Problem Formulation 15.2 Selection of Factors 15.3 Obtaining Random Samples of Paper Towels 15.4 How will the Absorbency be measured?
More informationHow To Run Statistical Tests in Excel
How To Run Statistical Tests in Excel Microsoft Excel is your best tool for storing and manipulating data, calculating basic descriptive statistics such as means and standard deviations, and conducting
More informationHow to set the main menu of STATA to default factory settings standards
University of Pretoria Data analysis for evaluation studies Examples in STATA version 11 List of data sets b1.dta (To be created by students in class) fp1.xls (To be provided to students) fp1.txt (To be
More information1.5 Oneway Analysis of Variance
Statistics: Rosie Cornish. 200. 1.5 Oneway Analysis of Variance 1 Introduction Oneway analysis of variance (ANOVA) is used to compare several means. This method is often used in scientific or medical experiments
More informationPoisson Models for Count Data
Chapter 4 Poisson Models for Count Data In this chapter we study loglinear models for count data under the assumption of a Poisson error structure. These models have many applications, not only to the
More informationOrdinal Regression. Chapter
Ordinal Regression Chapter 4 Many variables of interest are ordinal. That is, you can rank the values, but the real distance between categories is unknown. Diseases are graded on scales from least severe
More informationLinear MixedEffects Modeling in SPSS: An Introduction to the MIXED Procedure
Technical report Linear MixedEffects Modeling in SPSS: An Introduction to the MIXED Procedure Table of contents Introduction................................................................ 1 Data preparation
More informationAssumptions. Assumptions of linear models. Boxplot. Data exploration. Apply to response variable. Apply to error terms from linear model
Assumptions Assumptions of linear models Apply to response variable within each group if predictor categorical Apply to error terms from linear model check by analysing residuals Normality Homogeneity
More informationAn introduction to using Microsoft Excel for quantitative data analysis
Contents An introduction to using Microsoft Excel for quantitative data analysis 1 Introduction... 1 2 Why use Excel?... 2 3 Quantitative data analysis tools in Excel... 3 4 Entering your data... 6 5 Preparing
More informationTutorial for proteome data analysis using the Perseus software platform
Tutorial for proteome data analysis using the Perseus software platform Laboratory of Mass Spectrometry, LNBio, CNPEM Tutorial version 1.0, January 2014. Note: This tutorial was written based on the information
More informationOverview of Violations of the Basic Assumptions in the Classical Normal Linear Regression Model
Overview of Violations of the Basic Assumptions in the Classical Normal Linear Regression Model 1 September 004 A. Introduction and assumptions The classical normal linear regression model can be written
More informationAdditional sources Compilation of sources: http://lrs.ed.uiuc.edu/tseportal/datacollectionmethodologies/jintselink/tselink.htm
Mgt 540 Research Methods Data Analysis 1 Additional sources Compilation of sources: http://lrs.ed.uiuc.edu/tseportal/datacollectionmethodologies/jintselink/tselink.htm http://web.utk.edu/~dap/random/order/start.htm
More informationLinear Models in STATA and ANOVA
Session 4 Linear Models in STATA and ANOVA Page Strengths of Linear Relationships 42 A Note on NonLinear Relationships 44 Multiple Linear Regression 45 Removal of Variables 48 Independent Samples
More informationDESCRIPTIVE STATISTICS AND EXPLORATORY DATA ANALYSIS
DESCRIPTIVE STATISTICS AND EXPLORATORY DATA ANALYSIS SEEMA JAGGI Indian Agricultural Statistics Research Institute Library Avenue, New Delhi  110 012 seema@iasri.res.in 1. Descriptive Statistics Statistics
More information13. Poisson Regression Analysis
136 Poisson Regression Analysis 13. Poisson Regression Analysis We have so far considered situations where the outcome variable is numeric and Normally distributed, or binary. In clinical work one often
More informationStepwise Regression. Chapter 311. Introduction. Variable Selection Procedures. Forward (StepUp) Selection
Chapter 311 Introduction Often, theory and experience give only general direction as to which of a pool of candidate variables (including transformed variables) should be included in the regression model.
More informationCollege Readiness LINKING STUDY
College Readiness LINKING STUDY A Study of the Alignment of the RIT Scales of NWEA s MAP Assessments with the College Readiness Benchmarks of EXPLORE, PLAN, and ACT December 2011 (updated January 17, 2012)
More informationMultivariate Analysis of Variance (MANOVA)
Chapter 415 Multivariate Analysis of Variance (MANOVA) Introduction Multivariate analysis of variance (MANOVA) is an extension of common analysis of variance (ANOVA). In ANOVA, differences among various
More informationModule 5: Multiple Regression Analysis
Using Statistical Data Using to Make Statistical Decisions: Data Multiple to Make Regression Decisions Analysis Page 1 Module 5: Multiple Regression Analysis Tom Ilvento, University of Delaware, College
More informationReporting Statistics in Psychology
This document contains general guidelines for the reporting of statistics in psychology research. The details of statistical reporting vary slightly among different areas of science and also among different
More informationChapter 13 Introduction to Linear Regression and Correlation Analysis
Chapter 3 Student Lecture Notes 3 Chapter 3 Introduction to Linear Regression and Correlation Analsis Fall 2006 Fundamentals of Business Statistics Chapter Goals To understand the methods for displaing
More informationINTRODUCTORY STATISTICS
INTRODUCTORY STATISTICS FIFTH EDITION Thomas H. Wonnacott University of Western Ontario Ronald J. Wonnacott University of Western Ontario WILEY JOHN WILEY & SONS New York Chichester Brisbane Toronto Singapore
More informationdata visualization and regression
data visualization and regression Sepal.Length 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 I. setosa I. versicolor I. virginica I. setosa I. versicolor I. virginica Species Species
More informationCHAPTER 8 EXAMPLES: MIXTURE MODELING WITH LONGITUDINAL DATA
Examples: Mixture Modeling With Longitudinal Data CHAPTER 8 EXAMPLES: MIXTURE MODELING WITH LONGITUDINAL DATA Mixture modeling refers to modeling with categorical latent variables that represent subpopulations
More informationFactor Analysis. Chapter 420. Introduction
Chapter 420 Introduction (FA) is an exploratory technique applied to a set of observed variables that seeks to find underlying factors (subsets of variables) from which the observed variables were generated.
More informationSPSS ADVANCED ANALYSIS WENDIANN SETHI SPRING 2011
SPSS ADVANCED ANALYSIS WENDIANN SETHI SPRING 2011 Statistical techniques to be covered Explore relationships among variables Correlation Regression/Multiple regression Logistic regression Factor analysis
More informationMISSING DATA TECHNIQUES WITH SAS. IDRE Statistical Consulting Group
MISSING DATA TECHNIQUES WITH SAS IDRE Statistical Consulting Group ROAD MAP FOR TODAY To discuss: 1. Commonly used techniques for handling missing data, focusing on multiple imputation 2. Issues that could
More informationExample: Credit card default, we may be more interested in predicting the probabilty of a default than classifying individuals as default or not.
Statistical Learning: Chapter 4 Classification 4.1 Introduction Supervised learning with a categorical (Qualitative) response Notation:  Feature vector X,  qualitative response Y, taking values in C
More informationSilvermine House Steenberg Office Park, Tokai 7945 Cape Town, South Africa Telephone: +27 21 702 4666 www.spsssa.com
SPSSSA Silvermine House Steenberg Office Park, Tokai 7945 Cape Town, South Africa Telephone: +27 21 702 4666 www.spsssa.com SPSSSA Training Brochure 2009 TABLE OF CONTENTS 1 SPSS TRAINING COURSES FOCUSING
More informationAn SPSS companion book. Basic Practice of Statistics
An SPSS companion book to Basic Practice of Statistics SPSS is owned by IBM. 6 th Edition. Basic Practice of Statistics 6 th Edition by David S. Moore, William I. Notz, Michael A. Flinger. Published by
More informationBasic Statistics and Data Analysis for Health Researchers from Foreign Countries
Basic Statistics and Data Analysis for Health Researchers from Foreign Countries Volkert Siersma siersma@sund.ku.dk The Research Unit for General Practice in Copenhagen Dias 1 Content Quantifying association
More informationMaking the MDM file. cannot
1 2 Making the MDM file. For any given set of variables, you only need to do this once. But if you need to add a variable, you will need to repeat the process. HLM cannot transform variables (e.g., a square
More informationLinda K. Muthén Bengt Muthén. Copyright 2008 Muthén & Muthén www.statmodel.com. Table Of Contents
Mplus Short Courses Topic 2 Regression Analysis, Eploratory Factor Analysis, Confirmatory Factor Analysis, And Structural Equation Modeling For Categorical, Censored, And Count Outcomes Linda K. Muthén
More informationHow To Check For Differences In The One Way Anova
MINITAB ASSISTANT WHITE PAPER This paper explains the research conducted by Minitab statisticians to develop the methods and data checks used in the Assistant in Minitab 17 Statistical Software. OneWay
More informationI L L I N O I S UNIVERSITY OF ILLINOIS AT URBANACHAMPAIGN
Beckman HLM Reading Group: Questions, Answers and Examples Carolyn J. Anderson Department of Educational Psychology I L L I N O I S UNIVERSITY OF ILLINOIS AT URBANACHAMPAIGN Linear Algebra Slide 1 of
More informationDescription. Textbook. Grading. Objective
EC151.02 Statistics for Business and Economics (MWF 8:008:50) Instructor: Chiu Yu Ko Office: 462D, 21 Campenalla Way Phone: 26093 Email: kocb@bc.edu Office Hours: by appointment Description This course
More information