How to Test Ratings And Reviews on Android (Complete Guide)

Ratings and reviews are the primary signal users rely on when deciding whether to install an app from Google Play. A high average score drives organic installs, while a low score can trigger a steep d

Why Ratings and Reviews Matter for Android Apps

Ratings and reviews are the primary signal users rely on when deciding whether to install an app from Google Play. A high average score drives organic installs, while a low score can trigger a steep drop‑off in conversion. Beyond acquisition, the review text contains actionable feedback about crashes, usability problems, missing features, and privacy concerns. When a rating‑related flow breaks—such as the in‑app review prompt failing to appear, the submit button being disabled, or the review text being lost—users experience friction that often leads to negative public feedback and a lower star rating. Therefore, verifying that the ratings and reviews subsystem works correctly under realistic conditions is a quality gate that directly impacts business metrics.

Core Components of the Ratings & Reviews Flow

Most Android apps expose ratings and reviews through one of three mechanisms:

1. Google Play In‑App Review API – launches a modal dialog managed by Play Services, allowing users to write and submit a review without leaving the app.
2. Custom web view – loads a hosted web page (often a Firebase Dynamic Link or a proprietary server) that mimics the Play Store review UI.
3. Deep link to Play Store – uses an intent with market://details?id= to open the store listing; the user then manually taps “Write a review”.

Each path shares common UI elements: a rating bar (usually 1‑5 stars), an optional text field for comments, a submit button, and handling of dialog results (success, cancellation, error). Underlying the UI are network calls to Google’s review endpoints, token authentication, and occasional fallback logic when Play Services is unavailable. Understanding these components helps you design tests that cover both the UI layer and the service layer.

Test Matrix: Happy Path, Error Paths, Edge Cases, Accessibility, Security/Privacy

This matrix gives you a concrete checklist that can be turned into test cases for both manual and automated execution. Each row isolates a single risk factor, making it easier to trace failures back to root causes.

Manual Testing Approach: Step‑by‑Step Guide

1. Prepare the test device

• Install the target APK via adb install -r app.apk.
• Ensure Google Play Services is up to date (adb shell pm com.google.android.gms).
• Clear app data (adb shell pm clear com.example.app) to start from a clean state.
• Optionally enable network throttling (adb shell netem add delay 200ms) to simulate flaky connections.

1. Validate Play Services availability

• Launch the Settings app → Apps → Google Play Services → Permissions → confirm “Nearby” and “Storage” are granted.
• Run adb shell dumpsys activity services | grep com.google.android.gms to see if the review service is registered.

1. Execute the happy‑path flow

• Navigate to the screen that triggers the review prompt (often a “Rate us” button in Settings or after a completed level).
• Observe the dialog: for In‑App Review, it should be a full‑screen modal with Play branding; for web view, check the URL bar; for deep link, confirm the Play Store opens.
• Interact with the rating bar: tap each star, verify visual feedback (filled star) and haptic feedback if enabled.
• Enter a comment in the text field, then rotate the device to landscape and back; ensure the text persists.
• Tap Submit. Look for a toast or snackbar indicating success.
• After submission, revisit the review trigger point in the UI may show “Rate us” button; it should now display “Thanks for rating!” or be disabled.

1. Inject error conditions

• Enable airplane mode, repeat steps 3‑4, and confirm an error toast appears without crashing.
• Simulate an outdated Play Services version by side‑loading an older com.google.android.gms APK (use adb install -d old.gms.apk). Verify the fallback path.
• Rapidly tap the rating bar 15 times, then submit; check logs (adb logcat | grep Review) to ensure only one network request is sent.

1. Exercise edge cases

• Paste a 5000‑character string (generated with python -c "print('a'*5000)") into the comment field and submit; watch for truncation or validation error.
• Enter only emojis (e.g., “👍😀🚀”) and submit; after submission, open the Play Store listing via a browser to confirm the emojis appear correctly.
• Reduce available memory via adb shell am set-process-limit com.example.app 4 and launch the flow; ensure the UI renders without an OOM crash.

1. Run accessibility checks

• Turn on TalkBack (Settings → Accessibility → TalkBack).
• Swipe to focus on each star; listen for announcements like “star 1, button, not selected”.
• Double‑tap to select a rating; confirm the state changes and is announced.
• Navigate to the comment field and submit button; verify they are reachable and activatable.
• Disable TalkBack, then enable “Font size → Large” and “Display size → Large”; verify no clipping.

1. Perform security/privacy spot checks

• Use mitmproxy to intercept HTTPS traffic: set device proxy to the proxy’s IP and port, install the mitmproxy CA certificate on the device.
• Attempt to submit a review containing a SQL injection string (' OR 1=1 --). Observe the server response; ensure the app does not crash and shows a generic error.
• Launch the review flow from a background service (start a service via adb shell am startservice) and confirm that no UI appears and no network request is made without explicit user action.

1. Clean up

• Clear app data again if you plan to test another variant.
• Reset network settings (adb shell resetprop net.tcp.default_init_rwnd 65535).

Following this manual routine ensures you touch every cell in the test matrix and observe the app’s behavior under realistic variations.

Automated Testing on Android: Tools and Frameworks

Automating ratings and reviews validation requires interacting with system dialogs that may reside outside your app’s process. The most reliable approach combines UIAutomator (for cross‑app interactions) with Espresso (for in‑app UI) and, when needed, the Google Play In‑App Review API test library.

Tool Comparison

A robust automated suite typically layers Espresso for the in‑app portion, UIAutomator to launch and dismiss the review dialog, and optionally Firebase Test Lab to run on a matrix of real devices.

Setting Up UIAutomator for Review Dialogs

Add the dependency in your build.gradle (module):

androidTestImplementation 'androidx.test.uiautomator:uiautomator:2.2.0'
androidTestUtil 'androidx.test:orchestrator:1.4.2'

Create a test class:

@RunWith(AndroidJUnit4.class)
public class ReviewFlowTest {

    private UiDevice device;

    @Before
    public void setUp() {
        device = UiDevice.getInstance(InstrumentationRegistry.getInstrumentation());
        // Start the app from launcher
        Context context = InstrumentationRegistry.getInstrumentation().getTargetContext();
        final Intent intent = context.getPackageManager()
                .getLaunchIntentForPackage(context.getPackageName());
        intent.addFlags(Intent.FLAG_ACTIVITY_CLEAR_TASK);
        context.startActivity(intent);
        device.wait(Until.hasObject(By.pkg(context.getPackageName()).depth(0)), 5000);
    }

    @Test
    public void testInAppReviewHappyPath() throws Exception {
        // Navigate to the screen that shows the rate button
        OnView(withId(R.id.btn_rate)).perform(click());

        // Expect the Play Store review dialog to appear (system UI)
        // The dialog title contains "Rate this app"
        UiObject reviewDialog = device.findObject(new UiSelector()
                .textContains("Rate this app")
                .className(android.widget.FrameLayout.class.getName()));
        assertTrue("Review dialog did not appear", reviewDialog.waitForExists(3000));

        // Set rating to 4 stars (assuming 5-star bar with IDs)
        for (int i = 1; i <= 4; i++) {
            UiObject star = device.findObject(new UiSelector()
                    .descriptionContains("star")
                    .index(i - 1));
            star.click();
        }

        // Enter comment
        UiObject commentField = device.findObject(new UiSelector()
                .className(android.widget.EditText.class.getName()));
        .class.getName()));
        commentField.setText("Great app!");

        // Click submit
        UiObject submitBtn = device.findObject(new UiSelector()
                .text("Submit")
                .className(android.widget.Button.class.getName()));
        submitBtn.click();

        // Verify success toast
        UiObject toast = device.findObject(new UiSelector()
                .className(android.widget.Toast.class.getName())
                .textContains("Thanks"));
        assertTrue("Success toast not shown", toast.waitForExists(3000));
    }
}

Explanation of key steps

• UiDevice.getInstance gives you a handle to the device’s UI layer, letting you query elements that belong to other packages (e.g., the Play Store review dialog).
• The test launches your app via its launcher intent, ensuring a clean task stack.
• After clicking the rate button, we wait for a system dialog whose title contains “Rate this app”. This works for both the In‑App Review modal (which is still a system‑owned window) and the Play Store deep link (which opens a separate task).
• Rating assignment loops through star descriptors; many apps use contentDescription="star" for accessibility, making this approach robust.
• The comment field is located by class name; if your app uses a custom view, replace the selector with the appropriate resource ID or text.
• Finally, we assert that a toast with gratitude appears, confirming the review flow completed without error.

Espresso Test for In‑App UI

If your app uses a custom review screen (not the system dialog), Espresso suffices:

@RunWith(AndroidJUnit4.class)
public class CustomReviewScreenTest {

    @Rule
    public ActivityTestRule<ReviewActivity> activityRule =
            new ActivityTestRule<>(ReviewActivity.class, true, false);

    @Test
    public void submitReview_withValidData_showsSuccess() {
        activityRule.launchActivity(new Intent());

        onView(withId(R.id.ratingBar)).perform(setRating(4));
        onView(withId(R.id.editComment)).perform(typeView("Works well"));
        onView(withId(R.id.btnSubmit)).perform(click());

        onView(withId(R.id.txtSuccess))
                .check(matches(withText("Thank you for your feedback!")));
    }
}

Here setRating is a custom ViewAction that iterates over the rating bar’s stars and calls perform(click()) on each.

Running on Firebase Test Lab

Package your UIAutomator tests:

./gradlew connectedAndroidTest   # generates APKs in app/build/outputs/apk/androidTest/debug

Upload to Test Lab:

gcloud firebase test android run \
    --type instrumentation \
    --app app-debug.apk \
    --test app-debug-androidTest.apk \
    --device model=Pixel3,version=30,locale=en,orientation=portrait \
    --device model=PixelXL,version=28,locale=en,orientation=landscape

Test Lab will execute the tests on real hardware, capture logs, screenshots, and video, giving you confidence that the review flow works across diverse device configurations.

Code Samples: Instrumented Tests with Espresso/UIAutomator

Below is a complete, copy‑paste‑ready Gradle module that demonstrates both Espresso and UIAutomator tests for a typical ratings and reviews flow.

build.gradle (module)

plugins {
    id 'com.android.application'
    id 'org.jetbrains.kotlin.android'
}

android {
    compileSdk 34

    defaultConfig {
        applicationId "com.example.reviewdemo"
        minSdk 21
        targetSdk 34
        versionCode 1
        versionName "1.0"

        testInstrumentationRunner "androidx.test.runner.AndroidJUnitRunner"
    }

    buildTypes {
        release {
            minifyEnabled false
            proguardFiles getDefaultProguardFile('proguard-android-optimize.txt'),
                    'proguard-rules.pro'
        }
    }
}

dependencies {
    implementation "androidx.core:core-ktx:1.13.0"
    implementation "androidx.appcompat:appcompat:1.7.0"
    implementation "com.google.android.material:material:1.12.0"
    implementation "androidx.constraintlayout:constraintlayout:2.1.4"

    // Testing
    androidTestImplementation "androidx.test:core:1.5.0"
    androidTestImplementation "androidx.test.ext:junit:1.1.5"
    androidTestImplementation "androidx.test.espresso:espresso-core:3.5.1"
    androidTestImplementation "androidx.test.uiautomator:uiautomator:2.2.0"
    androidTestUtil "androidx.test:orchestrator:1.4.2"
}

Espresso Test (CustomReviewScreenTest.kt)

package com.example.reviewdemo

import androidx.test.espresso.Espresso.onView
import androidx.test.espresso.action.ViewActions.*
import androidx.test.espresso.assertion.ViewAssertions.matches
import androidx.test.espresso.matcher.ViewMatchers.*
import androidx.test.ext.junit.rules.ActivityScenarioRule
import androidx.test.ext.junit.runners.AndroidJUnit4
import org.junit.Rule
import org.junit.Test
import org.junit.runner.RunWith

@RunWith(AndroidJUnit4::class)
class CustomReviewScreenTest {

    @Rule
    @JvmField
    val activityRule = ActivityScenarioRule(ReviewActivity::class.java)

    @Test
    fun `submit review with valid data shows success`() {
        // Assume RatingBar uses the default Material rating bar
        onView(withId(R.id.ratingBar))
                .perform(actionWithId(R.id.ratingBar, setRating(4f))) // custom action below

        onView(withId(R.id.editComment))
                .perform(typeText("Great app!"), closeSoftKeyboard())

        onView(withId(R.id.btnSubmit))
                .perform(click())

        onView(withId(R.id.txtSuccess))
                .check(matches(withText("Thanks for your rating!")))
    }

    /** Helper to set a rating on a RatingBar */
    private fun actionWithId(viewId: Int, rating: Float): ViewAction {
        return object : ViewAction {
            override fun getConstraints() = MatcherMatchers.isAssignableFrom(RatingBar::class.java)
            override fun getDescription() = "Set rating to $rating"
            override fun perform(uiController: UiController?, view: View?) {
                view as? RatingBar ?: return
                view.isIndicator = false
                view.rating = rating
            }
        }
    }
}

UIAutomator Test (ReviewFlowUiAutomatorTest.java)

package com.example.reviewdemo;

import androidx.test.ext.junit.runners.AndroidJUnit4;
import androidx.test.platform.app.InstrumentationRegistry;
import androidx.test.uiautomator.UiDevice;
import androidx.test.uiautomator.UiObject;
import androidx.test.uiautomator.UiSelector;
import org.junit.Before;
import org.junit.Test;
import org.junit.runner.RunWith;

import static org.junit.Assert.assertTrue;

@RunWith(AndroidJUnit4.class)
public class ReviewFlowUiAutomatorTest {

    private UiDevice device;

    @Before
    public void setUp() {
        device = UiDevice.getInstance(InstrumentationRegistry.getInstrumentation());
        // Launch app
        Context context = InstrumentationRegistry.getInstrumentation().getTargetContext();
        Intent intent = context.getPackageManager()
                .getLaunchIntentForPackage(context.getPackageName());
        intent.addFlags(Intent.FLAG_ACTIVITY_CLEAR_TASK);
        context.startActivity(intent);
        device.wait(Until.hasObject(By.pkg(context.getPackageName()).depth(0)), 5000);
    }

    @Test
    public void testInAppReviewFallbackToPlayStore() throws Exception {
        // Trigger review flow (assumes a button with id btn_rate)
        OnView(withId(R.id.btn_rate)).perform(click());

        // Wait for either the in‑app review dialog or the Play Store activity
        UiObject reviewDialog = device.findObject(new UiSelector()
                .textContains("Rate this app")
                .className(android.widget.FrameLayout.class.getName()));
        boolean inAppShown = reviewDialog.waitForExists(3000);

        if (!inAppShown) {
            // Fallback: check that Play Store activity is in foreground
            UiObject playStore = device.findObject(new UiSelector()
                    .packageName("com.android.vending")
                    .className(android.widget.FrameLayout.class.getName()));
            assertTrue("Play Store did not open as fallback", playStore.waitForExists(5000));
        } else {
            // Proceed with in‑app review interaction (same as happy‑path test)
            // ... (rating, comment, submit) omitted for brevity
        }
    }
}

These snippets illustrate how to:

• Abstract rating setting for Espresso via a custom ViewAction.
• Launch the app and wait for its main UI before interacting with system dialogs via UIAutomator.
• Distinguish between the In‑App Review dialog and the Play Store fallback using package names and text selectors.

You can extend the UIAutomator test to cover error scenarios (airplane mode, outdated Play Services) by inserting adb shell svc wifi disable or installing an older com.google.android.gms APK before the test runs.

Autonomous, Persona‑Driven Exploration with SUSA

Traditional scripted tests excel at verifying known paths, but they often miss emergent issues that appear only when real users behave unpredictably. SUSA (susatest.com) addresses this gap by autonomously exploring an app using a set of modeled personas—each persona encapsulates a distinct behavior profile (e.g., “impatient” taps rapidly, “elderly” uses larger touch targets, “adversarial” attempts malformed inputs, “accessibility” relies on TalkBack, “power user” exploits shortcuts, etc.).

When you point SUSA at an APK or a web URL, it:

1. Installs the app on a fleet of real or emulated devices.
2. Selects a persona and drives the UI according to that persona’s policy (tap duration, scroll speed, input length, error‑recovery tolerance).
3. Triggers the ratings and reviews flow wherever it appears—whether via a floating action button, a settings menu, or a contextual prompt after a completed task.
4. Observes outcomes: crash logs, ANR traces, UI freezes, accessibility warnings, network failures, and unexpected state transitions.
5. Records each explored screen as a node in a graph; subsequent runs avoid revisiting known dead ends and prioritize unexplored edges, making the exploration progressively smarter.

Because SUSA does not rely on pre‑written test cases, it can discover bugs such as:

• A race condition where the review dialog is dismissed if the user rotates the device within 300 ms of opening it—a timing window that scripted tests rarely hit.
• An accessibility regression where TalkBack fails to announce the star rating after a font‑size change, only noticeable when the “elderly” persona (which forces large fonts) navigates the screen.
• A security oversight where an “adversarial” persona submits a review containing a JavaScript snippet; the app incorrectly renders it in a WebView, leading to a potential XSS vector.
• A privacy leak where the “curious” persona, after submitting a review, inadvertently triggers a background upload of the device ID to an analytics endpoint—detected via network traffic inspection.

SUSA’s output includes a PASS/FAIL verdict for each discovered flow, a video capture of the session, and a structured bug report (steps, logs, device state). Moreover, after each run it can auto‑generates Appium scripts and Playwright (Web)**—allowing you to turn the discovered by‑regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression regression